Dunbar’s Number matters online too

 Posted by (Visited 6701 times)  Misc  Tagged with: , ,
Feb 272009

Of course, this is completely unsurprising to me, since we demonstrated it via datamining of MMORPG metrics five years ago. There’s some interesting stuff here about “core” or tight-cluster friends versus the extended network, however.

The rise of online social networks, with their troves of data, might shed some light on these matters. So The Economist asked Cameron Marlow, the “in-house sociologist” at Facebook, to crunch some numbers. Dr Marlow found that the average number of “friends” in a Facebook network is 120, consistent with Dr Dunbar’s hypothesis, and that women tend to have somewhat more than men. But the range is large, and some people have networks numbering more than 500, so the hypothesis cannot yet be regarded as proven.

What also struck Dr Marlow, however, was that the number of people on an individual’s friend list with whom he (or she) frequently interacts is remarkably small and stable. The more “active” or intimate the interaction, the smaller and more stable the group.

— The size of social networks | Primates on Facebook | The Economist.

As someone with a larger-than-normal extended network and a smaller-than-normal core network, I kind of live with this every day as I use social media. There’s a lot of talk about the issue of “unbalanced” followers/following number on Twitter, for example, or about whether social media are used as marketing tools by some folks. In my case, the answer is undoubtedly “yes,” though perhaps my style of personal marketing is fairly informal. At the same time, as I have commented to folks at the office, the first anonymous brown-paper-wrapped package you get at your home address, first death threat, first random fan phone call at dinner, completely changes your perspective on social media…

  9 Responses to “Dunbar’s Number matters online too”

  1. I actually think that Dunbar and his infamous number gets misquoted and misunderstood more often than any other research of its type. It’s not a magic number, and taken out of context (like the FB sociologist did here) it’s almost meaningless.

    People often miss a few key points:
    * The derivation of the “Dunbar number” is based field studies on primate group behavior and Dunbar’s hypothesis that there is a correlation between relative neocortex size and group size.
    * Dunbar extrapolates from the measured primate data and the comparative size of the human cortex to reach a number of 147.8.
    * Dunbar extends his analysis to cultural and historical data that reinforces the “average” number of 150 for group size to include armies, nomadic tribes, terrorists, etc.
    * The number applies to groups with strong incentives to stay closely connected such as survival.
    * In order to maintain group cohesion, 42% of a person’s time must be spent performing “social grooming”, else the tenants of unstructured trust will not hold and the group will lose cohesion and group “rules” will not be followed, etc. And a hint from Dunbar as to how to address that dilemma: ” My suggestion, then, is that language evolved as a “cheap” form of social grooming, so enabling the ancestral humans to maintain the cohesion of the unusually large groups demanded by the particular conditions they faced at the time.”

    Christopher Allen (Life with Alacrity) extended Dunbar’s paper in 2004 describing why he thought there was a misunderstanding of Dunbar’s ideas based on a preoccupation with the absolute. From a modern world perspective and using social network analysis, Chris hypothesizes that that different group sizes impact a group’s behavior and their choice of processes and tools. Based on empirical data from MMOG and online communities, he suggests that for non-survival groups, the equivalent Dunbar number falls somewhere between 60-90.

    This comment is already too long – I wrote a post about this pertaining to the design of SL groups if you are interested.

  2. What also struck Dr Marlow, however, was that the number of people on an individual’s friend list with whom he (or she) frequently interacts is remarkably small and stable. The more “active” or intimate the interaction, the smaller and more stable the group.

    How is something like this that surprising? I’ve been saying for quite a long time now that the size of the network(graph) is inversely proportional to the strength of the ties(edges) that bind each individual member(vertices). I think there’s a post in the Mud-Dev archives somewhere where we debated and talked about this when someone brought up challenging Dunbar’s number law as presented in the Laws of Online Worlds.

    I always thought it was fairly common knowledge that your stable group of friends are your closest and they are the ones you keep in touch with the most. Being both your closest and the one’s you touch the most serves as a self fulfilling feedback loop of that inner circle. Active/intimate requires effort to maintain and that in and of itself limits the number that can be maintained that way.

    The 120 number of Facebook is considerably lower than the 250 quoted in the Laws. My guess is that FAcebook makes keeping in touch more difficult than say Twitter. I’d guess that the number of nodes in a Twitter network would be higher than that of a Facebook network and that the ratio is a simple measurement of the effort required to use the two applications for the task of maintaining a social network.

  3. Derek, I’m not sure if any importance needs to be placed on the 120 number vs the 250 number in the laws. I think that’s probably normal variance (with the added caveat of the below paragraph) attributable to the fact that we have insufficient data points. So rather than get hung up on the specific number, I think it’s more accurate to simply say that communities over a certain value, somewhere in the range of 1-200, give or take, begin to fragment.

    I’m also not really sure that FB communities are any harder to maintain than twitter ones, and I’d actually state that they’re way easier to maintain than virtual world ones. But remember, the measure of “number of FB friends” doesn’t actually require anything more than recognizing someone from real life and adding them to your friends list. No further commitment is required. So it’s more a measure of real life friends that are on FB, or friends of real life friends that are on Facebook, rather than it is of the maximum stable size of a developing community. Virtual World numbers on the other hand, tend to be more along the lines of relative strangers banding together, or at the very least being added to a core of people who know each other, so they come more from the developing community angle and might post bigger numbers.

  4. [insert obligatory snark about Marlow’s name]

    [insert obligatory reference to Ross Mayfield’s model]

    [insert obligatory reference to Christopher Allen’s points about groups breaking down]

    I’m actually thinking about developing a simulator to model relationships and a generic community, using things like Dunbar’s number, preferential attachment, and decay by neglect. Might be interesting.

  5. @Eolirin: Treat the graph of your friends network as a weighted graph where the edge weight is representative of the intimacy of the connection between two nodes. You could even go further and create the same graph as a weighted directed graph. That’s where I arrive at my conclusions in the above comment. What I said though is a SWHAG based on the small bits of data I’ve had available to me over the years from MUDs to social web sites.

    One can support more intimate friends by either having the natural ability to expend the effort required or by utilizing methods and practices that reduce the effort required on a per connection basis. Treat effort as a finite resource used to maintain a social network and quatify it along the edges of the graph. Of course quantifying the idea that Twitter requires less effort to maintain a connection is where things get difficult. What constitutes a unit of effort? That’s hard to say so all this is merely speculation until we can get some way of measuring this stuff in a large game.

  6. @Michael Chui: Thanks for that reference. I had no idea Christopher had did this work. I can only maintain a following on about three or four blogs while still maintaining my own. That page and his followup articles were good reading.

  7. Heh, similarly, I should probably always remember to link my work on this in any post related to the topic. 🙂


  8. Just to be helpful to potential readers, the best informal explanation of the Dunbar research I’ve found is here:


    Yes, it’s from Cracked.com. To be clear, I’m not joking here.

  9. Two items from other research into group dynamics and sociopathy:

    1. The oldest links are the most powerful and that is similar to what Derek is saying. That data came from a study of law firms, not social networks per se.

    2. Dunbar’s research also mentions grooming behavior as a means of creating the smaller stable inner clusters.

    Studying Facebook behavior is most interesting when studying messaging frequency among the nodes and assigning that the coupling strength, and as a mean to uncover hidden messaging going between connectors who are singleton members of multiple groups in competition in some way. It is a song of Ruth.

Sorry, the comment form is closed at this time.