What social media influence isn’t

Bernardo Huberman’s much-tweeted recently published study reveals that what makes people influential on Twitter is decidedly not their follower count. Based on analysis of 22 million tweets, the study looks at what factors correlate most closely with the spread of ideas as represented by links. Some celebrities and institutions with many followers are effective at getting pickup, and others aren’t. And some with not that many followers have influence well beyond the sphere of their own immediate network connections.

This disproves one of the basic illusion of social media douchebaggery – that by increasing one’s follower count, one will somehow gain in actual fame and fortune. This illusion justified many irritating techniques to gain followers, and more irritating boastfulness about the number of followers. Savvy and humane social media participants including Tara Hunt and Deanna Zandt have been talking about the truth that follower count doesn’t equal influence for a long time, and it’s finally visible in numbers.

But the study also reveals something less appealing, which is that when you start to focus attention on influence as the spread of links, then that metric becomes easily gameable. Huberman’s paper includes a list of Twitter handles that have outsize influence compared to their follower numbers. About half of these are actual people who are somehow good at spreading ideas though they are not personally popular, and half are contests rewarding the spread of links on Twitter.

Depending on the design and participation in the contest, this could mean that the contest is actually good at spreading ideas, or it could mean that it effectively incents people to click a software button to spread a link with with minimal connection to the content. Now, even when link-sharing is sincere and not just an empty game, people often have mixed motivations. In the course of ordinary social media interaction, a person may share a link to do a small favor to the original poster, to associate themselves with a cool person or topic, seek the attention of the poster and the community reading the post – many social motivations that are tangentially related to the content they are forwarding. Contests that reward mindless clicking are at the far end of a continuum of motivations for sharing content.

What Huberman’s results mean, though, is that by focusing attention on retweets and link-sharing as a primary measure of influence, that visible metric becomes subject to gaming. You can’t simply identify influence with retweeting, since calling attention to the metric can invoke gaming that makes the metric less meaningful.

This result has a number of implications for social software design. It reinforces what science communities have long known – that citations are a powerful measure of the influence of ideas; popularity contests, not so much. It also reinforces what game designers, economists, and business managers have known for a long time – people are motivated by what is measured, and publishing measurements changes the behavior that is measured. The trend and temptation endemic to social software design, to make invisible properties of the social network more visible, is not a simple act of measurement, but changes what is being measured. Those changes or may not be for the better.

18 thoughts on “What social media influence isn’t”

  12. Great analysis–and this obviously happened on the web when people understood the power of links and link -exchange–although set against a background of Web thinkers arguing that hyperlinking was good (irrespective of page rank).

    But I think you stop your analysis early. You assume that people designing systems to understand authority and impact in social contexts do so with naive assumptions that people won’t try to game the system.

    As designers of one such system, we are thinking a few steps beyond this–in order to find the right balance of metrics that are useful and a system that doesn’t lend itself to easy gaming.

    Finally, don’t forget that citation analysis (or citation counting) is no panacea: http://www.dlib.org/dlib/march09/canos/03canos.html And ironically, the undiscovered maths of pagerank–which despite its value to citation analysis, lay dormant for decades.

  17. Excellent use of “douchebaggery”! 🙂 IYHO, how close is this kind of link analysis to the kind I would do analyzing a botnet or a network of dubious characters? (Fairly similar, I’m guessing…)

Leave a Reply

Your email address will not be published. Required fields are marked *