You live in a network, a complex web of relationships that connects you with your family, your friends, your coworkers, and your classmates from your early school days. Some of these people you know within a close group, connected to you in a dense network in which everyone knows everyone else. Others, such as a friend you met by chance while on vacation, connect to you only loosely; she has close networks of her own, but you do not know them. We learn from early childhood how to navigate these relationships, figuring out to whom we must speak formally and with whom we can relax, who knows the best jokes and who needs our help. This network of relationships is the fundamental structure of society (Wellman 1999).1 Understanding these networks is essential for social scientists, who seek to better model how society functions; for activists, who want to change it; and for individuals, who need to navigate a rapidly changing and fragmented social world in which our network structures are becoming both larger and more diffuse than in the past.
If we could see a map of all these connections, stretching around the globe, we would see that a giant network of acquaintances links almost the entire world. Hundreds of years ago, this was not true; there were many isolated tribes and villages.2 But a century of mass migrations, missionaries, airplanes, road-building, and telecommunications have woven the world into a single social web. Today, with the possible exception of undiscovered remote tribes, these chains of acquaintanceship connect the entire human population. A series of historical social network maps would show us a profound change in how society is structured.
One reason to map social networks is to help us understand the implications of such a change. Is bigger better? Do technologically enhanced networks lessen the social network differences between urban and rural dwellers or exacerbate them? Mapping these changing networks helps the social scientist understand how the changes in society affect the way people form groups, receive support, and get information.
But network maps are not just a tool for social scientists and policy makers. They can be a guide for the members of this expanding society, helping them keep track of the increasingly numerous and complex connections among their acquaintances. Many people today live very mobile lives, changing cities and jobs frequently. Each move requires starting over in a new community, learning the relationships among new acquaintances from scratch. What sort of map would make these transitions easier? As the structure of our society moves to a scale “beyond being there,” network maps are an invaluable tool for making sense of our complex and diffuse web of connections.
Social network maps can potentially be part of the interface people use for communicating with each other. If I have 20 close friends, 200 acquaintances, an extended network of 1,200, and another few thousand very loose ties, how do I pick the subset of people I want to say something to, or who I would like to hear from at a particular time? Laboriously going through alphabetical lists of names is not useful; instead, I would prefer to be able to create groups around meaningful network clusters.
Another reason to map social networks is to create a data portrait; the map of your personal network is itself a portrait of you as a social being. Upon meeting a new person, a common conversational topic is “do you know so-and-so?,” as we try to place each other within our personal social world of friends and colleagues. Would seeing a network portrait of people we meet help us get to know them better? How would we manage such a map: what information is private or public? What should the map immediately display or slowly reveal?
In this chapter, we will examine conceptual issues in mapping social networks. This chapter is not a technical guide for producing these maps,3 but we will discuss the problem of who should be included and where should they placed. Legibility is a key issue in designing network maps, for a visualization of their interwoven connections can quickly become an unintelligible tangle. We will look at what meaning one can infer from the basic map and then explore approaches to depicting the information and support that flows through the network. We will talk about where the data come from, particularly online sources, and the challenges with this, such as incompleteness and maintaining privacy.
In theory, it is possible to draw a map of the network connecting all the people on earth. What would such a map actually look like? With a node for every person and a line connecting all acquaintances, it would be a hopelessly tangled mess. In fact, even when we reduce the scale quite a bit—say to the network consisting of you, your ties, and all of their ties—it is still a big and complex structure. Yet at this scale we can think about some practical questions. How do we decide where each person should go on the map? How do we want to define a connection: Is it everyone we’ve ever met over a lifetime? People we’ve been in contact with in the last five years? People to whom we would send a holiday card? You can sketch out a map of your own connections as a way to think about these problems. Take a piece of paper and sketch a social map of 50 or 100 or 400 of your friends, family, coworkers, and acquaintances. How did you cluster people? Whom did you put nearest yourself?
You may have started with your immediate family. It may be a cohesive group or fractured by feuds, divorce, and death. You may be very close to some members, while other relatives are emotionally distant. You may have included people you have worked with. Some jobs engender strong relationships, such as those between a student and mentor, company cofounders, and employees bonding in an exciting new venture or in an unhappy workplace. Or maybe work is just a place, and your coworkers only slight acquaintances. You may have worked at the same business for decades, or your social map may have several clusters of friends from various past positions. Some relationships span multiple parts of your life, such as a high-school classmate who is also in your running club and married to your cousin; this is someone you see at reunions, family get-togethers, and every other morning at 6:00 a.m. Other relationships are more tenuous, such as the people you are tied to via a single common connection, such as in-laws or a coworker’s friends. Some friends on your map you may see daily; others are far away and, though you feel attached to them, you may not have seen them for years. You may have included people who are neither friends nor family, but who have had a big impact on you, for example someone who caused an accident or saved a life. This social map is an autobiography. You know the stories of how you met each of these people and of your relationship with them, and how it has ebbed and flowed over time.
A hand-drawn map like this is idiosyncratic. Not only does it show your social world, it shows how you think about it. You may or may not have shown the relationships your acquaintances have with each other. You may have indicated different types of relationships: family versus work, romantic versus platonic. The categories you chose are part of your subjective view.
The metaphor of relationships as distance also shapes many social network maps: “He is a very close friend”; “I haven’t seen him for a long time”; “She’s a distant cousin.” Putting people who are close socially near each other on the map makes intuitive sense. If the depiction shows only your relationship with the other individuals, the map is possible to draw; you could, for example, make concentric circles of increasing intimacy. But once relationships among all the people are factored in, there are too many dimensions to place everyone ideally.
In drawing this map, you are likely to have encountered some of the difficulties that beset any attempt to depict social space. If you used distance to show the closeness of all the ties among the people, you probably quickly noticed that a two-dimensional graph cannot accurately depict the multiple constraints. You probably put people into groups, based on the context in which you know them, where they are, or their role in your life. But many people belong to multiple groups, so how do you show this?
The structure you are likely to have used is called an “egocentric” map. In the world of social network mapping, “egocentric” does not imply narcissism; it simply means a map that shows the connections of a single person. The map you drew consists of only your knowledge about certain people and their relationships. There may be friendships among them that you are unaware of. Other people, even a close friend who knows many of the same people, would depict these relationships differently.
The structure of the nodes and links in a network map highlights some interesting things. For example, if you draw people so that they are close to the others they know, you may see that there are “bridge figures,” people who connect otherwise separate groups (see figure 4.1). Bridge figures are conduits for information to flow from one group to another (Burt 2002; Granovetter 1973; Lin 2002). Because this is a map of your social world drawn from your perspective, you are the bridge between all the separate groups in it. Your network might be such that you are never the sole bridge between groups; if you live in the same town you grew up in, and socialize with friends from work, you will see mostly overlapping groups with multiple interconnections. But if you have moved around a lot, and have acquaintances from very separate social arenas (you may work in a law firm by day, practice salsa dancing at night, and do volunteer teaching for a month each summer in remote Australia), you will see a number of disconnected groups, linked only through you (see figure 4.2).
Since you know all the people in this sketch, you know that there is much more to the story of their connections. Some are truly close friends who would do anything for each other. Some share many of the same interests and beliefs, while others have conflicting ones. A gay rights activist and a religious fundamentalist may know each other, but they may never speak to each other. Or perhaps they are connected through some other dimension—maybe they are siblings or coworkers—and manage a cordial relationship through careful editing of what they discuss. You also know more about the individual personalities, such as who is gossipy, generous, or needy.
The network sketch of your acquaintances is a subjective view of a tiny piece of the vast network of human society. It includes only people you know, remember, and choose to include. It includes only a fraction of the information you have about the people and their relationships. This subjectivity is inherent to any social network depiction (Wellman and Wortley 1990). Whether the data come from exhaustive interviews by trained sociologists or from analyzing email records or social network sites, it will inevitably have many omissions. The sources’ perspectives will shape it, and the way “connection” is defined will limit it arbitrarily. Even so, social network maps are a very useful tool for making sense of people as social beings and for understanding the connections that shape society.
The map you drew used your memory of your acquaintances as its data source. It is an egocentric map, showing your extended network as seen from your perspective. Sociologists studying social networks collect this information by administering surveys and interviewing people.4 One challenge with this work is standardizing the information gathered from multiple people. Asked to name our “closest ties,” you might think of the one or two people you have relied on the most and for the longest time, whereas I might enumerate everyone I discuss anything personal with; our networks might actually be quite similar, but to such a survey they would seem quite different because of our different interpretations of the question (Bernard et al. 1990).
Researchers address this by asking more specific questions, such as to list everyone to whom they sent a Christmas card (Hill and Dunbar 2003) or from whom they would borrow $100. The boundaries of one’s network are always fuzzy. It’s clear that you know your sister and your best friend, and that the person who just walked past is a stranger. But what about the dry-cleaner whose shop you’ve visited every week for years? The sociologists’ questions and definitions create the boundaries of the networks they depict.
Another approach looks at “sociocentric” rather than “egocentric” networks. Instead of defining the network as the people connected to an individual, the sociocentric approach focuses on a limited populace—all the people who live in a neighborhood, work at a company, or publish in a field—and attempts to enumerate all the connections among the people in that group. Here the goal is to map the overall network structure, measuring density, finding key bridge figures, and so on (see, e.g., Wellman and Wortley 1990; Barabasi et al. 2002). For example, one might map the social networks of several departments in a company to understand how they access information and allocate responsibilities (Krackhardt and Brass 1994; Krackhardt 1990).
These days much information about social networks comes from online sources, such as email and social networking sites. The advantage here is that there is a great deal of easy-to-obtain data. One can write a program that goes through all of someone’s email (Adamic and Adar 2005) or SMS buddy lists (see figure 4.3; D’Angelo 2003), traces connections between blogs (see figure 4.4; Adamic and Glance 2005) or on a social network site (Ellison, Steinfield, and Lampe 2007; Heer and Boyd 2005), or monitors phone calls (Eagle, Pentland, and Lazer 2009) and extracts a network of relationships. Unlike interviews, which favor depth of information about a few contacts, electronic analysis generally provides a little information about a lot of contacts.5
The key problem here is in understanding the significance of the material. Electronic data mining measures a set of communicative acts, rather than emotions and impressions. It measures it impartially, but the data may be incomplete and unrepresentative. For example, a woman may be very close to her husband but email him infrequently, since they live in the same house. She may have the most email contact with her assistant at work, but that does not mean they have a strong tie, either personally or professionally. Even looking at the content of the messages can be misleading; the few emails she exchanges with her husband may be terse and factual, not because they are at odds, but because they use that medium mainly for logistical planning.
Some very social people send few emails because they prefer talking on the phone or sending text messages; other infrequent email users may be solitary loners who communicate little in any medium. Similarly, people use social network sites differently; some connect to only a few colleagues, others to thousands of strangers. Online communication provides a rich set of data for analyzing and mapping social networks, but it is important to keep in mind that each dataset provides only a partial and probably skewed view of its subjects’ social connections.
From private record keeping to political provocation, a social network map has many possible uses. With personal maps, the audience is familiar with the network; it is his or her own set of acquaintances. Here the goal may be to reveal unexpected connections or to provide a useful tool for navigating one’s social world. With other maps, the goal is to make the audience care about and understand the relationships in the depicted community. A map also makes it easier to explore a new territory. We can easily understand this in terms of physical travel; without a map, our progress is slow and tentative. Social maps have an analogous function. They can help us navigate new social territories quickly and they can help us manage an unprecedented number of social connections.
A personal network map can help people understand the relationships among their own acquaintances. Figure 4.5 shows all the people connected to one person (me) in the social network site Facebook in the summer of 2009. Any observer can quickly see the network’s size and basic clusters and structure.6
To me, it is a familiar landscape that I recognize and understand, though I had not previously seen it laid out in two dimensions. It is very much like looking at a map of a well-known neighborhood: it is a top-down abstraction that clarifies how everything fits together. I can easily recognize and interpret the clusters; the disconnected group at the upper right consists of friends from high school, the other small, tight cluster is a close-knit email group of women who had babies in August 1997.
This is an egocentric map; everyone depicted in it is connected to me. But it also includes information beyond my personal knowledge. Facebook allows you (usually) to see the connections of people with whom you are connected. This map includes that data—information that I am privy to on that site, but may not be aware of—and thus I can discover unknown connections. For example, I see that a European colleague knows the mother of one of my daughter’s friends; it turns out they are friends from their college days.
The map is interactive: clicking on a node shows the name of that person and highlights his or her connections within this network. As I write this, Facebook’s interface for browsing one’s friend list is a clunky alphabetical list, which shows nothing of the relationships among people and is dull to peruse. The network map is a far more beguiling way to look through one’s social neighborhood. A map like this could easily function as an interface, where clicking on a name would bring you to the person’s profile. Since it is situated in a site where people send messages, comment on each other’s updates and photos, play games, and so on, there is a wealth of social data that could be depicted in such a map.
In a world of small, tightly knit communities, you would not need these kinds of maps. Your social world would be limited enough, and repeated interactions frequent enough, that you could keep identities and relationships clear in your mind. Today, our social networks are expanding, owing to physical and socioeconomic mobility and to new communication technologies. In the past, we shed acquaintances as well as added them, but new social tools make this shedding less of an inevitability. Previously, keeping in touch was costlier, and fading weak ties—elementary schoolmates, acquaintances met at distant conferences, a friend’s cousin met at a wedding—would have dissolved entirely. Technologies such as Facebook make it easy to stay in casual contact with large numbers of people, and one of the primary attractions of social network sites has been to revivify such dormant ties (Smith 2011). The combination of mobile lifestyles that give us the opportunity to meet many people and social technologies that make it easy for us to stay in touch is creating enormous active personal networks, too big to manage without assistance. They are social landscapes in need of a map.
A map of one’s personal social network can function as a portrait.
People post maps of their own networks to photo-sharing sites, much as they post photographs of themselves and their friends at various events; the point is to show who you are, what you do, and whom you know (see figure 4.3). Some are annotated, explaining the significance of the clusters to others (see figure 4.6).
Like any portrait, the network map shows only one perspective of its subject. Although the map of my network in figure 4.5 includes many people, it is far from a complete—or even representative—sample of the people I know. It shows the people I know who happened to be on Facebook back in 2009 and who had chosen to connect to me and vice versa. It does not include some of my closest friends, my children, my neighbors, or my dentist, and it does include some people I have a very tenuous relationship with: vaguely remembered high-school classmates and colleagues met at a long-ago conference. Furthermore, many of the people who appear to have only a few connections actually have very large networks of friends on the site, but I am not friends with their friends so they do not appear on this map.7
The selection of who is included in any network depiction is somewhat arbitrary. The Facebook networks vary by how people use the technology: some connect only to close friends, a subset of the people they would send holiday cards to; some connect to anyone who asks; and many gather thousands of on-site ties, few of whom they would recognize if they ran into them on the street. The visualization accurately portrays how the subject uses that particular technology. But one must be careful in making broader assumptions about the subject’s sociability or personality since his use of a particular technology may not be typical of his overall behavior.
The map of a friend’s network can be fascinating to peruse. The names are familiar, and thus exploring the relationships is interesting (see figure 4.7). A similar map of a stranger’s connections provides some, but limited, insight. One can see if that community is unusually large or small, and whether it is a dense network of interconnected ties or a loose network of many separate groups. If it is a map of a community, rather than an egocentric map of an individual’s connections, one can see the individual’s role: Does she have many connections? How well connected are her connections? Are they densely connected among themselves, or is she the bridge between diverse groups? (See this chapter’s appendix for more on bridge figures and network structure.)
A social network map has the potential to be a vivid and evocative portrait, but the information must be more than just the skeleton shape of the network. Most contemporary network maps are just nodes and generic connections. To tell a story, we need more information than that. Which are the close connections? What brought these people together? What is surprising about finding this person next to that one?
A network map can be a narrative device, providing the framework for a detailed annotation of people and their relationships.
Dramatis Personae (see figure 4.8) is a map of the social connections among the guests at a wedding. It is a hand-drawn map that describes itself as “a useful (if inexhaustive) visual taxonomy of the binding ties … between the characters gathered here today.” The purpose of this map was to introduce the guests at a wedding, providing them with some background about the various relationships to help start conversations. Had it simply connected names with lines like a typical network map does, it would be a nice image, but hardly a compelling conversation starter. However, the mapmakers annotated each line with a pithy comment about the relationship between the two connected people: “Mick, father of the bride learned to play bridge from the mother of Bill.” “Andy nearly died of sunburn after swimming in the Elk River with Chrissy.” Here, the network map functions as the foundation for telling stories. While a few of the annotations provide basic relationship information—“June is the proud mother of Officiant Ellen”—most are notes about intriguing exploits or quirky facts: “Kay makes the favorite strawberry jam of Andy.” It is network map as humorous ice-breaker.
Any organization—a club, a corporate or academic department—could collectively create a similar map as a way to capture its social history. One can imagine this being especially helpful to newcomers as they attempt to learn the often taken-for-granted and unspoken structure of the community.
Other social network maps tell stories to critique society. Mark Lombardi made hand-drawn network maps that depicted the relationships among people and corporations in a variety of complex scandals: the relationship between big oil companies and terrorists, the collapse of the Vatican bank, and so on (see figure 4.9). The data they depict, though newsworthy, could make for a dense and dry textual narrative. But drawn as elegant maps, they have been exhibited, reprinted, and widely discussed.
Lombardi’s drawings are activist maps, telling stories of undue influence and compromising relationships. Here, the hand-drawn quality conveys subjectivity; it reminds the viewer that an artist is behind the map, and the map shows the world through his eyes.
Drawing a social map by hand provides freedom that an automated map does not have. Computer-rendered maps are limited to the expressive range of the rendering software (which can be quite limited or very powerful), whereas making a map by hand can provide greater flexibility in how a line is drawn or which symbols are used. Lombardi depicted the nuances of social structure using arrows, circles, and scribbles. His diagrams still greatly simplify complex relationships, but they are more informative than simple line-and-node network maps (Tufte 2006).
Hand-drawn maps can be subjective, even propagandistic. Computer-generated maps generally depict some mathematical relationship derived from a dataset, and although one can certainly use biased or fictional data as the basis, the map is a faithful depiction of that data. The hand-drawn map is more easily skewed; here, where each node and link is individually marked, accuracy is harder and drawing by feel and intuition easier. That said, the hand-drawn map conveys to the viewer that it is a personal expression, whereas the computer-rendered one appears more objective, “data-driven,” and authoritative, whether or not it actually is.
Unlike hand-drawn maps, where the process of individually marking each node and link limits the size of the network it is practical to depict, computers can render immense datasets of connections. Here the challenge is to make the densely complex image legible to the human viewer.
Figure 4.10 shows Jeffrey Heer’s map of three degrees of his connections (his friends, their friends, and the friends of those friends) on Friendster, an early social network site. It has 47,471 nodes with 432,430 connections among them. Brightness represents degree: Heer himself is the brightest node (0 degree), followed by his connections, and so on. The image is striking, but also illegible.
The impenetrable denseness of this image, however, is part of its message.8 Friendster and several other early social networking sites were designed around the idea that if you could follow a chain of connections to someone, a chain of a length that sounds small, like three, four, or five hops, you had a real connection to that person. In reality, that is a big social distance. You are probably at most four hops from everyone in your city (and less than that in a smaller town), but that does not mean the stranger you just passed is nearly an acquaintance. The tangle of Heer’s map makes the immensity of the three-degree distance palpable.
We usually, however, want something more comprehensible. One of the most powerful ways of bringing legibility to a dense dataset is to make the visualization interactive (Ahlberg, Williamson, and Shneiderman 1992; Becker, Eick, and Wilks 1995; Heer, Bostock, and Ogievetsky 2010; Keim 2001; Shneiderman and Aris 2006; Yee et al. 2001). This allows the viewer to explore the data, uncovering different patterns. With network visualizations, a common simplification is to remove nodes—either before the layout is made (thus simplifying the dataset that is shown) or after (thus removing some of the clutter from the image, but leaving the remaining nodes as they were located within the full dataset of connections). By alternating between the full view, which shows overall structure and clusters, and the filtered view, which reveals details such as individual connections, the viewer gains a fuller picture of the structure (Ahlberg and Shneiderman 1994; Jia et al. 2008; Kumar and Garland 2006).
Visual Who, which we discussed in chapter 1, used both interactivity and filtering. The viewer explored the community’s relationships by adding and removing anchors (representing foci of interest) to attract different names of community members, based on their affinities. The names of people with strong ties to the current anchors would be brighter and layered on top of those with a weaker affiliation, effectively filtering the display (see figures 1.1 and 1.2).
The lines in a social network map represent relationships ranging from vague acquaintances to the closest of relatives. Information may flow symmetrically between them, or only one way. The next stage in designing social network maps involves looking more closely at what flows through the network.
The basic network of nodes and links is the skeleton of society, setting its fundamental shape. Mapping the flow of information (or money, services, social support, etc.) through this network gives us a much fuller picture of the significance of the connections.
A map that shows nodes and links is a map of potential flow—of support, gossip, germs, or the like. However, because something can flow through a connection does not mean that it necessarily does—a road map shows where you may drive, but it does not tell you where traffic flows lightly or is jammed up with bumper-to-bumper traffic (Zuckerman 2008). In a social network, information can potentially flow from any person who possesses it to anyone with whom they connect; but in practice, of course, we do not tell everything we know to everyone we know. The network maps we have looked at so far show people and their connections as generic nodes and links. But, in terms of information flow, people are filters, modifiers, amplifiers, and sources. We do not share all the information we have with everyone we know, and when we do, we change it from the form in which we received it. We add commentary and omit details. Sometimes we attenuate it because the story is unimportant or inappropriate to pass on. Sometimes we amplify it, adding remarks or credibility that will make it more likely that others too will pass it on. Even if Beth repeats verbatim to Charlie something she heard from Alex, the message is different because the source is different: Alex may be more believable than Beth, or something that sounds ordinary coming from Alex is quite surprising to hear from Beth. Moving through the network transforms information.
Breaking bin Laden (see figure 4.11) by Gilad Lotan is a visualization of the spread on the Twitter network of the news that Osama bin Laden had been killed. This image and Lotan’s discussion of the event illustrate the complex relationship between connections, trust, and value in determining how information flows through a network (Lotan 2011).
On Sunday evening, May 1, 2011, the White House announced that President Obama would be making a special address to the nation later that night. This prompted much speculation on Twitter about what the subject might be, with guesses ranging from news about Libya’s Gaddafi to various jokes. Although some speculated that it involved the death or capture of bin Laden, none of these tweets had a big impact until Keith Urbahn, a Twitter user with a relatively modest following, posted “So I’m told by a reputable person they have killed Osama bin Laden. Hot damn.” Hundreds of people almost immediately reposted or responded to this tweet.
Urbahn was not the first to mention bin Laden in this context. There was other speculation, but those postings did not spread. What was different about Urbahn’s? For one, Urbahn was a trustworthy source: he had been Chief of Staff for former Defense Secretary Rumsfeld, a fact that most repostings of his message mentioned. People saw this source information as significant, for on Twitter, where each post is limited to 140 characters, such identifying information is included only when essential.9 Furthermore, his post was declarative and decisive sounding. Others had made similar predictions, but more tentatively. For example, a security expert with a larger following than Urbahn’s had written, “I’m saying OBL is dead—I want to be the first on Twitter to engage in complete speculation that might be correct.” Although this writer had expertise and connections, he presented only a guess and his posting made little impact (Lotan 2011). Urbahn’s post, citing a “reputable person,” was worded as fact.10 Network position also helped. Among Urbahn’s relatively small following were some influential people. Brian Stelter of the New York Times wrote to his more than 50,000 followers: “Chief of staff for former defense sec. Rumsfeld, @keithurbahn, tweets: ‘I’m told by a reputable person they have killed Osama bin Laden.’” And finally, it was exciting yet believable news. Other predictions were less attention getting (the announcement will be about the death toll in recent storms) or too far-fetched (the president will announce that a comet is heading toward Earth) (US Political Madness Forum 2011). Here we see how in a social network, the role and reputation of the people, the network’s structure, and the content of a message shape information flow.
Why spread such a story? Part of the dynamic was social support. People were intrigued and concerned about the mysterious announcement. Chatting on Twitter or other forums was a way to share the anxiety and try to find out some news as quickly as possible. But there is also an element of status competition. In the information-based world of Twitter, being at the forefront of a new story—being one of the earliest posters of a story that becomes big—confers prestige. And on Twitter, prestige immediately translates into increased influence: if more people follow you, your subsequent postings reach a bigger audience. Urbahn’s list of followers went from about 300 to over 5,000 in the twelve hours following his bin Laden posting (Gavin 2011) and rose to 8,000 in the following days.
Being the first to convey a piece of information is valuable. If you hear something novel and fascinating from me, that raises my standing in your opinion. But being at the forefront is also risky. Announcing a story before it is verified makes you a leader, but if you turn out to be wrong, you lose credibility; do so often enough, and no one will listen to anything you say. Yet if you wait to say anything until the information you are passing on is well established, you gain little prestige, and indeed risk boring your listeners. The tension between the rewards of being first and the risk of being inaccurate is one of the key forces shaping the dynamics of public, information-based social networks.
Having a public voice has only recently, with the rise of mass online communication, become commonplace. Forums such as Twitter, blogs, and other social network sites form a middle ground between personal communication and large-scale broadcast, such as TV and newspapers. They have the personal connections of traditional conversation—I am likely to be more influenced by something a friend tells me than a stranger—but their reach, and the speed at which news travels through a network, is far greater. Mapping how information flows through these new channels can help us understand how these technologies are changing society.11
Although networks of personal connections form the basis of our communities, their structure can be hard to see even in familiar circumstances. Mapping how information flows in our everyday life lets us perceive the dynamics that shape our communities.
An elementary school, with its grades, classes, and ever-changing friendships, is a small but complex social world. To help students see the intricate web of their relationships and understand how this affected who learned what from whom, Rick Borovoy and his colleagues created i-balls, simple animations and games played on small keychain devices. Children could create, play, and—most interestingly—pass these games on to others by linking their devices together. The goal of the project was to recreate, in a form that could be recorded and visualized, the richly innovative and sociable world of children’s folklore, where stories and traditions are invented, refined, and spread from person to person (Borovoy et al. 2001).
In one version, the designers gave out the devices to the students and staff at a public elementary school. The project was very popular, and over the course of a few weeks they designed and disseminated hundreds of i-balls. The students were fascinated by maps showing how the i-balls traveled (see figure 4.12). They could see which ones had circulated mainly within a single grade, and which had spread beyond a limited group, and who was the bridge making that connection. They could see who shared which game or picture widely, and who received it but chose not to pass it on. “There was a sense of excitement around this privileged view these students were getting of a geometry they always sensed, but could never before directly apprehend” (Borovoy et al. 2001, 470).
The bin Laden Twitter visualization and the i-ball maps show how information—a rumor, a game—spread throughout a community. Sometimes, however, the act of communication itself is what is interesting, and thus we want to see which links in a network are active or dormant.
In the mid- to late 2000s, the most popular social network site was MySpace, with well over 100 million users. It was a bustling, diverse, and sometimes gritty virtual city, populated with regular users keeping in touch with friends, bands and celebrities promoting their new projects, as well as spammers and other undesirable accounts.
One problem users faced was forming an impression of strangers they encountered on the site. Unlike Facebook, which discouraged people from linking to strangers (and ultimately greatly surpassed MySpace in size and influence), MySpace encouraged people to add connections, and it was quite common to get numerous friend requests from strangers. Some might be intriguing—someone with common interests or who had posted interesting content. But not every friend request was desirable. Some were from “collectors”—people trying to accumulate as many “friends” as possible, who often used automated programs to solicit connections. Worse, an attractive-looking profile might be a front for a spam account that would fill one’s profile with unwelcome advertising (Zinman and Donath 2007).
Another problem was simply making sense of the activity on the site. Which of your friends was active? Whom were they communicating with? Like other early social network sites, MySpace had no news feed (a feature Facebook introduced in 2006, which gave users a continuous stream showing what their friends were posting)—one had to visit each individual profile to see what, if anything, was new.
Comment Flow* by Dietmar Offenhuber is a network map annotated with snippets of conversation, designed to bring social legibility to such sites (see figure 4.13). It shows the rhythm and volume of interchanges between people and provides a glimpse of their content, using as source data the public comments people posted on their connections’ profiles. By showing a bit about a person’s relationships with his various connections, this annotated social network diagram creates a more nuanced data portrait (Offenhuber and Donath 2008). One gets a very different impression from seeing a person with numerous friends, most of whom appear to be other people, all having various sorts of social conversations, than one gets from seeing a person with an equivalent number of connections, most of whom are promotional entities, their “conversation” a one-way stream of press releases and announcements.
Mapping information flow and revealing hidden patterns can make for a vivid portrait or community mirror; it can also violate the privacy of the depicted individuals. We need to be cognizant, when designing or displaying these renderings, of people’s expectations of what is public or private. The display of data visualizations should reflect the publicness of the underlying data.
Email, for instance, is generally private. Different levels of information can be derived from it: a network map may simply show a person’s contacts; it may show how much communication occurs through the different connections; or it can show what the people are saying to each other. I may feel comfortable allowing other people to see a map of who my contacts are, but not one that shows what we discuss. And, though you may think that one would never want to make the content of email public, there are situations where it could be appropriate. For example, a group of people working on a project, who know from the start that their communication is public, at least among themselves, could find such depictions useful.
Omitting identifying information, limiting words, and otherwise abstracting the data can make a visualization derived from private material appropriate to a public setting. The data used in Comment Flow were semipublic, more visible than a private email, but certainly meant for a smaller audience than the general public. For public settings, it could omit the message text and show just boxes to indicate the quantity and frequency of messages.
It is important to keep in mind that the data about a community may be unrepresentative of the actual relationships among the people. My closest friend may not be on the map at all if she does not use the site, or if I am linked to her, it may show no communication, though in fact we are in frequent contact—just via other media and in person. It may show lots of communication with someone who is not at all a close tie, but who is a prolific writer. Some of this information is outside the designer’s knowledge, such as how much do a pair of people communicate elsewhere. To clarify and provide context of the picture, additional data may be incorporated. Am I one of five ties Bob has, or one of seven hundred? Are Sue’s five messages to me typical of her correspondence rate, or am I the only person to whom she has written?
Much flows through a network besides information: social support, money, and so on. Sociologists gathering network data often ask questions designed to elicit information about the strength and function of each tie (Hogan, Carrasco, and Wellman 2007; Roberts et al. 2009). How often do you speak with this person? Is this a relative, a coworker, a lover? Would you confide in him or her about a relationship problem? Would you ask to borrow money from him or her, or vice versa? This is fascinating information, and certainly can turn the dry outlines of the network into a much more colorful narrative. When can these data be part of a community map? The surveying sociologist works on the condition of anonymity; his subjects appear in reports identified by pseudonyms: their friends do not see that they’ve been judged to be only a casual acquaintance, or an unreliable person to confide in. One display approach for these data is to aggregate and anonymize them to create community mirrors that do not identify specific individuals, but can still show us the types of communities that form in different parts of our country and how our peers live.
The artist Kelly Sherman used wedding reception seating plans to create a series of family portraits (see figures 4.14, 4.15). In these spare, stylized, and mostly gray graphics, the location of the bride and her parents is marked in red and that of the groom and his parents in yellow; these simple charts hint at complex family histories and dynamics. The tales are told in divergence from the norm. In most of the arrangements, the bride and groom sit at a table near, but not with, their parents, but in figure 4.14, they are together on a dais, evoking a close-knit family with strong ties between generations. In figure 4.15, the bride and groom are near the bride’s parents and the groom’s mother and stepfather, but the groom’s father and stepmother have been relegated to the other side of the room, separated by the wide expanse of band, buffet, and dance floor.
While these are not network maps, they are abstract—yet vivid—depictions of people’s relationships. As the viewer, we try to understand what makes some families diverge from the expected arrangement. Similarly, one goal in designing network maps is to help a community evolve its understanding of norms in order to see anomalies; this is what makes a network map truly legible.
Unlike the impression given by a network diagram’s uniform nodes and simple links, real social networks are composed of diverse individuals with complex motivations. They choose what to pass on and to whom, shaping the dynamics of flow in the network. And they continuously reshape the network itself. Sharing gossip and providing support to another makes a tie stronger. (Or, it usually does. Sharing highly critical opinions or political views another finds offensive can sever ties.)
Technology changes these dynamics. It makes communication easier, while also altering people’s motivations for communicating.
In our face-to-face conversations, we are selective about what we tell and to whom. If you hear a rumor that your boss is going to be replaced by someone from another department, to whom do you tell this? Not to everyone you know, because most would not care. You tell one colleague at work because she is a good friend and this change will affect her; but you don’t tell others there because you are unsure if it is true and do not want to be the source of misinformation. You have lunch with a friend from outside work and tell him, not because he is deeply interested, but because you are chatting about your day and how things are going at work. Throughout the day, you are constantly taking all the news you are privy to and knowledge you have accumulated and deciding whom to tell and why. You may have a new kitten, but you only tell funny kitten stories to other cat lovers. You may have heard an off-color joke; who else will find it funny? You do not insert your entire store of knowledge at random into conversations, but as the topics in a discussion change, you bring up related facts and anecdotes.
Publication media such as blogs and Twitter make it more efficient for us to communicate with many people at once, but they make it hard to tailor our communication for specific relationships. Thus, many of the things we read on such sites are of little interest; but we also become privy to stories that we otherwise would not have heard. At its worst, the overly wide audience is a privacy failure (we discuss the effect of such social context collapses in chapter 11, “Privacy and Public Space”). At its best, this new form of serendipitous learning can strengthen relationships, as when we come to see more aspects of a person we had known only in a limited context (which we discuss further in the next chapter, “Our Evolving Super-Networks”).
Figure 4.16 is a map of the world drawn entirely from data about the Facebook connections between people. Lines connect cities, and their brightness represents the number of Facebook ties between the two locations as of 2010. The map has no drawn geographic outlines; the recognizable world map emerges from the connection data alone. We began the previous chapter by noting that, were it possible to map all the connections among everyone on Earth, a historical series of such maps would reveal profound changes in society’s structure. The Facebook Map of Global Connections is one of the closest approximations we have to such a map.
This map shows a recognizable world, its geography resembling—but far from identical with—the familiar outlines of continents. It is similar to a population map—but with some interesting exceptions. Connections in Europe stop abruptly as it meets the Russian border, and the massive population of China produces barely a light. In Russia, a local site dominates social network activity, and China had blocked access to Facebook the year before the map was made. It is a map not of connections in general, but one that shows how communication flows—or not—through one increasingly influential medium.12
At another level, we can read many stories into this map: it shows the ties emigrants maintain with their families at home, college friendships kept long after the students graduate and scatter to distant jobs—the global web of relationships spun by a highly mobile population (Zuckerman 2013). It is not a route map: technically, any two places where there are Facebook users link as easily as any other pair. Instead, it shows the social relationship between places. Although figure 4.16 shows geographical contours, it actually depicts the collapse of distance.13
It also hints at the future. The development of any new channel—whether roads and air routes to travel on or new media to communicate with—affects the social structure of the linked communities. In the next chapter, we will look at how increased mobility—both social and geographical—changes the form and function of our social networks.
The basic graph of a social network, with the people as nodes and their connections as the edges between them, is an abstract and minimalist representation of the complexities of human social structures. Yet we can still glean insight about the roles and relationships in a community by analyzing a simple map of their connections.
One of the basic attributes of the nodes in a network is centrality. Centrality is a way of measuring a person’s role in a network—how likely is it that she will be aware of what is going on with the others? Does she control access to information? One can make a basic assessment of centrality with only the graph of a network, without knowing details of the characteristics of the ties.
Figure 4.17 is a simple social network, designed by network analyst David Krackhardt, to explain the network concept of “centrality” (Krackhardt 1990; Krebs 2004). There are several ways to measure how central a node is in a network; we will look at three of them: degree, betweenness, and closeness centrality.
Degree centrality is the most straightforward. It measures how many direct connections a node has. Someone with high degree centrality knows a lot of people, and someone with low degree centrality might be a newcomer, a loner, or someone peripheral to the group. In the network shown in figure 4.17, Don, who has the most connections, six out of a possible nine, has the highest degree centrality, whereas Jan has the lowest, with only one.
However, although having many connections can make one privy to much information, it is also important to measure how disparate one’s sources are. Although Don has the most connections, they are all within a tight cluster. Hiro, who has only three connections compared to Don’s six, has more diversity in his network; he thus has a higher betweenness centrality. Like Don, he has access to the information in the tight cluster on the left, but also to the separate world of Ike’s knowledge. Furthermore, Hiro is what is known as a bridge figure. He is the sole link between two (or more) groups. He controls what information moves between them, and were he to leave, they would be disconnected.
Another form of centrality is closeness, which measures how far you are, on average, to other members of the network. As ties become increasingly indirect, their value quickly fades. Here, Fay and Gina have the highest closeness centrality; Jan, off at a distance, has the lowest.
In this mini-network, we can compute these measures by hand, but once the network becomes more complex, it is easiest to do so computationally. Degree centrality is the number of connections someone has divided by the number of total possible connections. This network contains a total of ten people, so the most connections one could have is nine. Jan, with one connection, has a degree of 1/9, while Hiro, with three connections, has a degree of 3/9. Closeness centrality is measured by summing how far someone is from each member of the network (what is the shortest path between them) and dividing the total number of connections by that sum. Fay needs fourteen hops to reach each person in the network, giving her a closeness centrality of 0.643, whereas Jan, who is quite distant from most, has a closeness of 9/29 or 0.310. Betweenness centrality is measured by determining the shortest path between all the nodes and counting the number of times a node is in the midst of one of these paths. Here, Cara, Ed, and Jan are never part of a shortest path, so their centrality is zero.
In mapping social networks, we can use these measures to help viewers understand the community. We can highlight key figures, such as the popular member with high degree centrality, or the important bridge figures. We can use variations on these measures, depending on what story we want to tell: some take into account the varied strength of different ties, whereas others are optimized to understand a particular problem, such as how catastrophic for the community would it be to lose a particular member of the network.
There are numerous other network features to explore. Networks can be densely or loosely connected. The ties between people can be strong or weak, narrowly focused or diversely multiplexed; a tie might represent a formal work relationship within a clear hierarchy, a warmly supportive friendship, or a wary truce between ex-partners. Some of these features (such as density) can be represented with a minimal graph, whereas others require more nuanced information than the sparse node and connection model. A full overview of social network analysis techniques is outside the scope of this book; to learn more, see Borgatti et al. 2009; Garton, Haythornthwaite, and Wellman 1997; Girvan and Newman 2002; and for a history of the research field, see Freeman 2004.