A recent article came out in Forbes.com, “Is the Tipping Point Toast?” tackling popular assumptions about the role of influencers and opinion leaders in dissemination and adoption, particularly in the area of online/viral media. The article focuses on recent work by Duncan Watts (a physicist that has gained popular recognition for his publications on “small world theory” and his recent employment by Yahoo) in contrast with the works “The Tipping Point” (Malcolm Gladwell) and “The Influencers” (Ed Keller and Jon Berry). The crux of the article is Watts’ claim that really, “influencers” have no more effect on other peoples’ activities than the “average joe” in the network.

Since a lot of social media analysis is based on the assumption that it is possible to identify “opinion leaders” (influencers), and that the most effective PR and marketing strategies should be based upon building relationships with these opinion leaders, this entire discussion becomes very relevant to social media. Without the concept of structure and influence, marketing / PR must focus entirely on the content and chance. I believe that “influence” is demonstrable in these applications, but that we need to be careful that our usage of the term “influencers” does not get muddled with a lay concept of a few influential elite. Perhaps as social media analysis becomes more sophisticated, we can start measuring different types of influence and different social media community structures, so we can be precise in our understanding of adoption, propogation, influence, infection, and dissemination in a social media context.

Two years ago I did some markov-chain experiments on this subject to test how network structure and individual network characteristics can impact influence, propogation, and dissemination of something through a network. My experiment ran thousands of simulations on different types of random network structures (small world, scale-free / preferential attachment, scale-free / fitness model, etc.), with different network sizes and a handful of randomly-selected starting points. Each actor in the network also had a random “resistance” to new ideas (following conclusions in the classic work on diffusion theory, “Diffusion of Innovations” by Everett Rogers).

Here were some of my findings:

• Network structure strongly impacts widespread adoption. This is consistent with the work of Dr. Tom Valente (see “Network Models and the Diffusion of Innovations“), which makes a claim that cliques and subgroups have perhaps the greatest impact on diffusion.
• The density and size of the network affect adoption and dissemination. Denser networks will cause more “reinforcing” links among individuals.
• The centrality (connectiveness) of the starting individuals is statistically significant, but is related to the overall network structure. Closeness centrality (reach, connectedness) had significant impact (P<.001), more in uniform networks than in sparse or structured networks. Betweenness centrality (”gatekeepers” and “brokers”) had by far the most impact in all cases (P<.0001). My simulations showed that “average Joes” did not have the same effect for propogation and dissemination as “well-connected Joes” (influencers), except perhaps collectively.
• I discovered an interesting characteristic was Velocity - the speed by which an idea propogated through the network.
• Individual resistances are not that important relative to other characteristics (above).
• Results largely followed predictions by Rogers (see above), in the classic “S-curve” of adoption:

  

Valdis Krebs has a nice little graphic and summary on the different types of network influence determined by one’s place in the network, and the network structure on his post “Duncan Vs. The Influentials“. His post was also the motivation for my own discussion of this topic.

How one chooses to model a network is a critical determinant. In the past, Mr. Watts has used a “small world” model for randomly generating simulated networks that is structurally very different from another popular model proposed by another physicist (Alberto-Laszlo Barabasi) based on preferential connections. This may seem academic, but if one ignores network structure, then one will always have a distorted view of influence in that network. Given that Dr. Barabasi’s work was ground-breaking in how well it managed to explain large-scale networks - especially the internet - I think it’s critical to frame any discussion of viral media in terms of preferential-attachment and scale-free networks.

On a final note, I think that fertile ground for researching “influence”, especially regarding viral media, is Erdös and Rényi’s concept of “percolation”. In physics, percolation describes the process in which a sudden phase change takes place; for example, when water shifts from liquid to gas (boiling) or solid (ice). “…Before, we have a bunch of tiny isolated clusters of nodes, disparate groups of people that communicate only within the clusters. After, we have a giant cluster, joined by almost everybody.” (from Barabasi’s book). In this quote, the “giant cluster” is the group of individuals that may have shared and/or viewed a particular viral video or email, for example. Perhaps I will dig into this more in another post, but I also found percolation theory may deal with the topic of “fence sitters” as it describes how individual clusters of molecules may “waffle” back and forth between states until the phase shift takes place.

| Permalink | Trackback