Clay Shirky's Writings About the Internet:
Networks, Economics, and Culture

Communities, Audiences, and Scale.

April 6, 2002

Prior to the internet, the differences in communication between community and audience was largely enforced by media -- telephones were good for one-to-one conversations but bad for reaching large numbers quickly, while TV had the inverse set of characteristics. The internet bridged that divide, by providing a single medium that could be used to address either communities or audiences. Email can be used for conversations or broadcast, usenet newsgroups can support either group conversation or the broadcast of common documents, and so on. Most recently the rise of software for "The Writable Web", principally weblogs, is adding two-way features to the Web's largely one-way publishing model.

With such software, the obvious question is "Can we get the best of both worlds? Can we have a medium that spreads messages to a large audience, but also allows all the members of that audience to engage with one another like a single community?" The answer seems to be "No."

Communities are different than audiences in fundamental human ways, not merely technological ones. You cannot simply transform an audience into a community with technology, because they assume very different relationships between the sender and receiver of messages.

Though both are held together in some way by communication, an audience is typified by a one-way relationship between sender and receiver, and by the disconnection of its members from one another -- a one-to-many pattern. In a community, by contrast, people typically send and receive messages, and the members of a community are connected to one another, not just to some central outlet -- a many-to-many pattern [1]. The extreme positions for the two patterns might be visualized as a broadcast star where all the interaction is one-way from center to edge, vs. a ring where everyone is directly connected to everyone else without requiring a central hub.

As a result of these differences, communities have strong upper limits on size, while audiences can grow arbitrarily large. Put another way, the larger a group held together by communication grows, the more it must become like an audience -- largely disconnected and held together by communication traveling from center to edge -- because increasing the number of people in a group weakens communal connection.

The characteristics we associate with mass media are as much a product of the mass as the media. Because growth in group size alone is enough to turn a community into an audience, social software, no matter what its design, will never be able to create a group that is both large and densely interconnected.

Community Topology

This barrier to the growth of a single community is caused by the collision of social limits with the math of large groups: As group size grows, the number of connections required between people in the group exceeds human capacity to make or keep track of them all.

A community's members are interconnected, and a community in its extreme position is a "complete" network, where every connection that can be made is made. (Bob knows Carol, Ted, and Alice; Carol knows Bob, Ted, and Alice; and so on.) Dense interconnection is obviously the source of a community's value, but it also increases the effort that must be expended as the group grows. You can't join a community without entering into some sort of mutual relationship with at least some of its members, but because more members requires more connections, these coordination costs increase with group size.

For a new member to connect to an existing group in a complete fashion requires as many new connections as there are group members, so joining a community that has 5 members is much simpler than joining a community that has 50 members. Furthermore, this tradeoff between size and the ease of adding new members exists even if the group is not completely interconnected; maintaining any given density of connectedness becomes much harder as group size grows. As new members join, it creates either more effort or lowers the density of connectedness, or both, thus jeopardizing the interconnection that makes for community. [2]

As group size grows past any individual's ability to maintain connections to all members of a group, the density shrinks, and as the group grows very large (>10,000) the number of actual connections drops to less than 1% of the potential connections, even if each member of the group knows dozens of other members. Thus growth in size is enough to alter the fabric of connection that makes a community work. (Anyone who has seen a discussion group or mailing list grow quickly is familiar with this phenomenon.)

An audience, by contrast, has a very sparse set of connections, and requires no mutuality between members. Thus an audience has no coordination costs associated with growth, because each new member of an audience creates only a single one-way connection. You need to know Yahoo's address to join the Yahoo audience, but neither Yahoo nor any of its other users need to know anything about you. The disconnected quality of an audience that makes it possible for them to grow much (much) larger than a connected community can, because an audience can always exist at the minimum number of required connection (N connections for N users).

The Emergence of Audiences in Two-way Media

Prior to the internet, the outbound quality of mass media could be ascribed to technical limits -- TV had a one-way relationship to its audience because TV was a one-way medium. The growth of two-way media, however, shows that the audience pattern re-establishes itself in one way or another -- large mailing lists become read-only, online communities (eg. LambdaMOO, WELL, ECHO) eventually see their members agitate to stem the tide of newcomers, users of sites like slashdot see fewer of their posts accepted. [3]

If real group engagement is limited to groups numbering in the hundreds or even the thousands [4], then the asymmetry and disconnection that characterizes an audience will automatically appear as a group of people grows in size, as many-to-many becomes few-to-many and most of the communication passes from center to edge, not edge to center or edge to edge. Furthermore, the larger the group, the more significant this asymmetry and disconnection will become: any mailing list or weblog with 10,000 readers will be very sparsely connected, no matter how it is organized. (This sparse organization of the larger group can of course encompass smaller, more densely clustered communities.)

More Is Different

Meanwhile, there are 500 million people on the net, and the population is still growing. Anyone who wants to reach even ten thousand of those people will not know most of them, nor will most of them know one another. The community model is good for spreading messages through a relatively small and tight knit group, but bad for reaching a large and dispersed group, because the tradeoff between size and connectedness dampens message spread well below the numbers that can be addressed as an audience.

It's significant that the only two examples we have of truly massive community spread of messages on the internet -- email hoaxes and Outlook viruses -- rely on disabling the users' disinclination to forward widely, either by a social or technological trick. When something like All Your Base or OddTodd bursts on the scene, the moment of its arrival comes not when it spreads laterally from community to community, but when that lateral spread attracts the attention of a media outlet [5].

No matter what the technology, large groups are different than small groups, because they create social pressures against community organization that can't be trivially overcome. This is a pattern we have seen often, with mailing lists, BBSes, MUDs, usenet, and most recently with weblogs, the majority of which reach small and tightly knit groups, while a handful reach audiences numbering in the tens or even hundreds of thousands (e.g.

The inability of a single engaged community to grow past a certain size, irrespective of the technology, will mean that over time, barriers to community scale will cause a separation between media outlets that embrace the community model and stay small, and those that adopt the publishing model in order to accommodate growth. This is not to say that all media that address ten thousand or more people at once are identical; having a Letters to the Editor column changes a newspaper's relationship to its audience, even though most readers never write, most letters don't get published, and most readers don't read every letter.

Though it is tempting to think that we can somehow do away with the effects of mass media with new technology, the difficulty of reaching millions or even tens of thousands of people one community at a time is as much about human wiring as it is about network wiring. No matter how community minded a media outlet is, needing to reach a large group of people creates asymmetry and disconnection among that group -- turns them into an audience, in other words -- and there is no easy technological fix for that problem.

Like the leavening effects of Letters to the Editor, one of the design challenges for social software is in allowing groups to grow past the limitations of a single, densely interconnected community while preserving some possibility of shared purpose or participation, even though most members of that group will never actually interact with one another.


1. Defining community as a communicating group risks circularity by ignoring other, more passive uses of the term, as with "the community of retirees." Though there are several valid definitions of community that point to shared but latent characteristics, there is really no other word that describes a group of people actively engaged in some shared conversation or task, and infelicitous turns of phrase like 'engaged communicative group' are more narrowly accurate, but fail to capture the communal feeling that arises out of such engagement. For this analysis, 'community' is used as a term of art to refer to groups whose members actively communicate with one another. [Return]

2. The total number of possible connections in a group grows quadratically, because each member of a group must connect to every other member but themselves. In general, therefore, a group with N members has N x (N-1) connections, which is the same as N2 - N. If Carol and Ted knowing one another count as a single relationship, there are half as many relationships as connections, so the relevant number is (N2 - N)/2.

Because these numbers grow quadratically, every 10-fold increase in group size creates a 100-fold increase in possible connections; a group of ten has about a hundred possible connections (and half as many two-way relationships), a group of a hundred has about ten thousand connections, a thousand has about a million, and so on. The number of potential connections in a group passes a billion as group size grows past thirty thousand. [Return]

3. Slashdot is suffering from one of the common effects of community growth -- the uprising of users objecting to the control the editors exert over the site. Much of the commentary on this issue, both at slashdot and on similar sites such as kuro5hin, revolves around the twin themes of understanding that the owners and operators of slashdot can do whatever they like with the site, coupled with a surprisingly emotional sense of betrayal that the community control, in the form of moderation.

(More at kuro5hin and slashdot. [Return]

4. In Grooming, Gossip, and the Evolution of Language (ISBN 0674363361), the primatologist Robin Dunbar argues that humans are adapted for social group sizes of around 150 or less, a size that shows up in a number of traditional societies, as well as in present day groups such as the Hutterite religious communities. Dunbar argues that the human brain is optimized for keeping track of social relationships in groups small than 150, but not larger. [Return]

5. In The Tipping Point (ISBN 0316346624), Malcolm Gladwell detailed the surprising spread of Hush Puppies shoes in the mid the '90s, from their adoption by a group of cool kids in the East Village to a national phenomenon. The breakout moment came when Hush Puppies were adopted by fashion designers, with one designer going so far as to place a 25 foot inflatable Hush Puppy mascot on the roof of his boutique in LA. The cool kids got the attention of the fashion designers, but it was the fashion designers who got the attention of the world, by taking Hush Puppies beyond the communities in which it started and spreading them outwards to an audience that looked to the designers. [Return]

Thanks to NYU's Information Law Institute,, who commented on an earlier version of this argument, and to Jessica Hammer for editorial review.