Of late, the social blogosphere’s been much abuzz with talk of influence and social scoring systems. Yet many, including myself, have pointed out, there are clear failings in the single-number-of-influence approach: To name but a few, most single number metrics today take no account of offline influence, blog content, industry-specific influence and, yes, at least for Klout, they overvalue tweet frequency and volume in their weightings.
When it comes down to it, it seems that one of the core assumptions behind some of the single-number influencer scoring systems is that ‘bigger is better”. And, on the face of it, doesn’t that make sense? Consider Klout scores. Key components of that score reward more followers, more tweets, more tweets per minute, more re-tweets. When all these variables are combined via the the magical black box scoring algorithm, don’t we all see some logic in the rank ordering of the influential twitter accounts?
What's more important are Klout clients and agencies investing in Klout Perks programs getting their return? Are we going to expect the best results - the spread of influence - from rewarding high Klout scorers? I think marketers need to be aware of some lesser known data points before creating their strategy. So here we go...
First to be aware of is that there is a huge big numbers bias. Where I think this "big number" bias stems from is that influencer scores like Klout seem to adopt what influencer researchers call, "The Influentials Hypothesis", namely, the notion that the most connected nodes in a social network are , by definition, the most influential nodes. In layman's terms, there are specialized individuals in every population who tell us what to do.
When Big Numbers Fail
But is bigger always better in terms of actually influencing our online behavior? A tantalizing mini-test of this was posted by Megan Garber at Neiman Journalism Lab, describing a social influence experiment done by Nicholas Christakis in promoting his book, Connected. In order to drive sales, Christakis armed three “influential” twitterers with an endorsement tweet, including a link to Amazon to buy his book.
The celebrity actress’s tweet had no effect whatsoever. The well-known tech-book publisher achieved only 1 sale for his tweet. But in defiance in our belief in the power of large numbers, Susannah Fox, the Pew Internet Trust analyst, sold three copies of the book into her relatively scnat 4,906 follower base (measured at the time).
Yes, this is a very small and casual study. But the direction of the results is not in line at all with expectations for influence by either follower counts or with the Klout score of these individuals.
Lesson 1: Celebrity tweets (or just any user with large follower count’s tweets) may not help sales of an academically-oriented book on social networking.
Lesson 2: Interests seems to matter. Ms. Fox’s intellectual leanings and interests - which overlap the subject content of the book- may have something to do with her relatively better performance with book sales.
Is there more? Studying a database of over 500 million tweets, one Stanford study looked at the relation of twitter follow count to the ability to propagate hashtags. Users with large follower counts did not propagate as well those with smaller follower counts (<1000 followers). UPDATE: In another study of 90 million Twitter posts, of influence diffusion looking at high influentials and low influentials, the authors found "...under a wide range of plausible assumptions the most cost-effective performance can be realized using “ordinary influencers”—individuals who exert average or even less-than-average influence."
Could it possibly be that it is the little nodes in the social graph, the smaller networks out on the edges of the social graph, which direct strong ties of influence? With some marquis consumer brands paying up to $10,000 per endorsement tweet , perhaps the results of these (and hopefully, follow-on) studies deserve more attention.
Better View from an Interest Graph?
These studies suggest we need a more refined view of our social graphs and how influence is diffused. One attempt in the right direction, one promising to provide a deeper understanding of the role played by social network infrastructure, is provided by Brian Solis’ recent study of the interest graph for Starbucks top followers, using Research.ly and Peoplebrowsr. In this study of 50,000 twitter accounts, Solis was able to show patterns of like-interest groups among the Starbucks top followers, diving a level deeper into showing who these followers are, their shared topics and their relationships.
In the long run, perhaps it is not the single-number influence scoring systems that will help develop our knowledge of how influence works, but rather those emerging data analytic tools like Research.ly mPact and Blogdash which look at social data sets and track keyword usuage and the users who tweet them.
Above all, the studies hint to me that we need to cast off our inherent human fallability, our bias, that the large numbers in the top-level variables (eg. follower count,tweet frequency, amplification) within our current data view of The Social Graph are the primary drivers of influence.
Ironically, viewing the final Starbucks Interest graph produced, there still seems a strong visual bias to depict the large nodes, the large follower count followers, suggesting these "large node" users are driving the key interactions.
But what if , as in the Christakis social experiment and the Stanford study on hashtag propagation, the real driver of influence, namely the decision to purchase a latte/go with friends for a Starbucks cafe experience, is stemming from users with low follower counts. See those little nodes out on the outside of the social network? Are they having tighter engagement with their followers?
Here’s a crazy thought experiment to consider: If the “Susannah Fox” effect is real, doesn’t that suggest that, at least to increase book sales conversions, rather than use one high profile “influential” with 1 million followers, book conversions would do considerably better to pay 200 Susannah Foxes, each with some 5000 followers? If such moderately connected influentials could be identified (aye, that’s the rub), Nicholas Christakis might have sold 600 books with one tweet. Now that's a hypothesis worth testing. (And yes- I'm motivated to read Christakis' book now.)
Who Are The Influencees?
It's clear we are still in the discovery phase. We know too little about the nature of influence. The real problem I have with single-number influence metrics is that, while useful as online Q-scores for celebrity marketing deals, these numbers turn off our brains on thinking about influence. And that's dangerous for marketers leading an influence strategy toward ROI for clients. Our inherent human bias to seek shortcuts and easy solutions may well be holding us back from asking deeper questions. Far too often clients ask "How do we find our influencers?" when, as Christakis has pointed out, we might more pertently ask, "Who are our influencees?"
It turns out this viewpoint is a much less-travelled existing road in influence research, one which posits that it's "the influencees" or "the susceptibles" that we ought to be focusing on. One seminal 2006 study titled Influentials, Networks, and Public Opinion Formation used mathematical modelling to examine the dynamics of how influence could disseminate. A key finding of the paper is:
Large-scale changes in public opinion are not driven by highly influential people who influence everyone else but rather by easily influenced people influencing other easily influenced people.
Who are these highly influenced people? Interestingly, a 2009 Harvard study, Do Friends Influence Purchases in a Social Network? found that it was the moderately connected people, not the highly connected, that were the most likely likely to be influenced by friend's purchases.
While we're here, if you're not one of the 500,000 people who have already viewed it, Google researcher, Paul Adam's presentation on "The Real Life Social Network" makes some great points on how we might focus on smaller group structures within our larger social sphere to get better insight on influence.)
In the end, the secret to understanding the the still nebulous concept of influence, the recommendations and endorsements that really drive our actions, may lie in understanding the bonds within smaller networked groups of "susceptibles". (Obviously, this is nothing new to the influence researchers. However, marketing folks throwing large dollars for client companies with celebrity tweets ought to be reviewing the details of their strategy.)
Just maybe, fuelled by our addiction to the ease of one-number-influence scores, we’re attacking this problem upside down. Inconceivable as it may seem to us now --- maybe it’s not the activity within the low-to-medium Klout scorers, so-called "low influencers", but the activity of the high Klout scorers that is specious and distracting.
Take away: Maybe the marketing agencies and clients aiming to influence through Klout Perks ought to aim, not at the celebrities (including social media celebrities), but at the lesser knowns who are moderately connected and whose interest graphs centers on their product and/or service.
Image 1 credit: Carter Hodgkin