Duncan Watts and his latest study of Twitter data shows that consumers with average influence may offer a better return on marketing dollars.
Together with Eytan Bakshy, Winter Mason, and Jake Hofman, Watts examined the attributes and relative influence of 1.6 million Twitter users by tracking 74 million diffusion “events”, as tracked by retweets and reposts, that occurred over a two-month period in 2009.
Twitter “presents a promising natural laboratory for the study of diffusion processes,” they write. Since users “follow” the broadcasts of other users, multi-step diffusion processes can be reconstructed by crawling follower graphs. Further, the use of URL shorteners (e.g., bit.ly) allowed the researchers to track all diffusion patterns, not just those that were successful .
Not surprisingly, the Twitter dataset revealed that the largest cascades tended to be generated by users who were influential in the past and who had a large number of followers. However, most individuals with these attributes were not successful in generating large cascades. The vast majority of URLs did not spread at all and even moderately sized cascades were extremely rare.
In addition, a sampling of 1000 URLs showed that, while URLs that spread widely tended to be more interesting and elicited more positive feelings, content characteristics didn’t distinguish diffusion success from failure either.
Overall “predictions of which particular user or URL will generate large cascades are relatively unreliable,” they conclude. Thus, to consistently harness word of mouth influence, marketers need to target large numbers of potential influencers, thereby capturing average effects.
Another knock at Gladwell’s theory of influentials, but it is hard to argue with empirical work – there may be more value to be had in viewing each consumer as an influencer in his or her own right, with many small cascades of information sharing eventually adding up to marketing success.