On Twitter, Followers Don’t Equal Influence

9:30 AM Friday May 7, 2010
by Scott Berinato | Comments (5)

It could be that Twitter research is popular because Twitter data is free and so accessible. That’s okay. Gift horses are just as good for riding.

The best, latest entry in Twitter research is the handiwork of Meeyoung Cha from the Max Planck Institute for Software Systems in Germany. (Co-authors are: Hamed Haddadi, Royal Veterinary College, University of London; Fabricio Benevenuto, Federal University of Minas Gerais, Brazil and Krishna P. Bummadi, also from Max Planck Institute.)

Cha called her paper, “The Million Follower Fallacy,” a term that comes from work by Adi Avnit. Avnit posited that the number of followers of a Tweeter is largely meaningless, and Cha, after looking at data from all 52 million Twitter accounts (and, more closely, at the 6 million “active users”) seems to have proven Avnit right. “Popular users who have a high indegree [number of followers] are not necessarily influential in terms of spawning retweets or mentions,” she writes.

We asked Cha about the findings, published by the Association for the Advancement of Artificial Intelligence.

HBR: What impelled you to look at this?

Cha: There’s a lot of focus on Twitter’s follower count. Many think that having many followers will make their tweets spread more quickly and widely in the network. They try to increase their followers to increase their influence. Further, there are advertising companies that offer users tips on how to increase their follower count and they pay popular users with a lot of followers to insert ads in their tweets. We sought to investigate this hype in a more rigorous way. We thought that gathering a large amount of data from Twitter would give us a great opportunity to conduct data-driven social science.

Your results seem to suggest that number of followers does not equal influence and that other factors show that number of follows is in fact a bad indicator of actual influence. That would mean those companies both marketing techniques to increase followers, and the ones paying tweeters with large numbers of followers, are in the wrong business.

I think it would be too strong to claim that follower count is abad metric. Our claim is that follower count is not sufficient to capture the influence of a user (i.e., the ability of an user to sway the opinions of her followers). It only shows how popular the user is (i.e., the size of her audience). But, as we showed in our paper, retweets and mentions, which measure the audience responsiveness to a user’s tweets, do not correlate strongly with number of followers.

Now a very interesting question would be, “How should one measure influence?” We are afraid that there are no easy answers here. One would have to take a combination of many metrics, including follower count, mentions, and re-tweets. However the hard part is figuring out the relative importance of the component metrics. This is the subject of our future research.

What are the implications of your research to businesses — both the ones wondering how to use Twitter and the ones wondering what mentions of their brands on Twitter mean?

Businesses, rather than trying to put emphasis on the follower count, could try to increase audience responsiveness in their fields. We have done some work on investigating the roles of different types of users (e.g., mass media, small businesses, grassroots users) in spreading news topics. As you might have guessed, mass media played a significant role in spreading popular topics. But when it comes to non-popular or even niche topics, small businesses and opinion leaders were far more effective in engaging audience than mass media. This may be because mass media tend to talk only about popular topics (otherwise, they might be considered spammers).

Some of your conclusions seem obvious. For example, news organizations spawned influence on a broad array of topics while celebrities were good at getting their names retweeted. Are we just confirming what is intuitive here?

Good question. One could view the primary contribution of this particular piece of work as grounding or confirming the intuition or expectations people already have using real data. Though we also feel that had the conclusion been that follower count is the primary metric that matters, many people might have still felt that the findings are expected as well.

That said, this is just the first of a series of studies where we intend to use Twitter data to analyze how information (news, rumors, conventions) spread through out the social network. We expect some of these future studies to go beyond confirming existing theories and yield new (potentially surprising) findings.

It sounds like from some of the paper that Twitter works mostly as news distribution and a gossipy RSS feed. What do you think of the value of Twitter?

Like you mention, there is a cacophony of responses and replies to users’ tweets. The metric that captures this is mentions. However, we think Twitter is more than just a gossipy feed. Twitter is a nice content filter for identifying topics that engage millions of people. The act of retweeting (based on my personal experience) typically indicates that the receiver read the tweet carefully, found it interesting, and deemed it to be of sufficient interest and value to forward it further to her followers. In some sense, retweets capture the content value of the tweet. Retweeted messages are typically of interest to a larger audience as opposed to personal tweets that are primarily of interest to one-hop friends and never forwarded.

What was the most surprising finding to you? What did your research elicit that you didn’t expect?

We were surprised by how only a fraction of Twitter users actively tweet. And this small fraction of Twitter users provoke responses (mentions) and initiate information cascades (retweets). I guess many people use Twitter to browse others’ messages rather than generating a lot new messages themselves.