In the past week Gawker made the startling claim that the controversial conservative politician Newt Gingrich has 93% fake followers on Twitter. Backstory: Gingrich himself had complained why so little attention had been given to the fact that he has such a large Twitter audience:
“I have six times as many Twitter followers as all the other (Conservative) candidates combined, but it didn’t count because if it counted I’d still be a candidate; since I can’t be a candidate that can’t count.”
Gingrich has a whopping 1,325,842 followers. Twitter is increasingly being regarded as a proxy indicator of popularity and influence.
But, said Gawker, an ex-staffer told them that:
“Newt employs a variety of agencies whose sole purpose is to procure Twitter followers for people who are shallow/insecure/unpopular enough to pay for them.”
Later Gawker followed that up with some research. They had found a research firm PeekYou that professes to have scrubbed all of Newt’s followers. They claim to use sophisticated tecniques to check names against actual people on the web. And only 8% were deemed to be ‘real’ people.
We let our own RAAK bots loose to see what we could find out about Newt’s followers. In the space of a day, they gathered a random sample of some 26,616 followers. And here is what we got.
We also had a look at the dates the accounts were created, and these were evenly distributed. There were no flurries of activity.
None of these numbers seem to be suggesting anything as dramatic as PeekYou’s assertion of 92% fake accounts.
Meanwhile, The Center for Complex Networks and Systems Research at Indiana University’s School of Informatics and Computing has analyzed a random sample of 5,000 of Gingrich’s followers. Their data is strikingly similar to ours. Fully a third had never posted to Twitter and 76% had added no biographical information to their profile.
So what does our and the center’s data mean?
It’s a widely known fact that Twitter is full of lurkers. People that sign up only to follow celebrities but that never make their presence felt. Telling these people apart from fake accounts can be a daunting task. For example, this and this account following Newt could be by lurkers. They are following a coherent list of accounts, which would sit pretty well with the idea that this is a person that just wants to consume content.
And it gets more complicated. Even accounts that Tweet can be non human. We ourselves have created bots that Tweet and that have attracted large follower counts; even from what we can tell are real people (as opposed to other bots). One of our bots even had a Klout score of over 50!
In another recent analysis, we tracked these weird accounts that tweeted political Google News results. They had avatars, bios and were starting to pick up followers. They turned out to be bots, and we presumed that they were being used as sock puppets. Heaven knows for what purpose. Shortly after we started tracking them, they must have been found out and Twitter deleted them.
So none of the above stats of ours conclusively point to whether an account is fake or not. The absence of a bio or Tweets only increase the likelihood.
A better way to guess is to compare the preponderance of these suspicious accounts amongst the Twitter population as a whole with those of Newt. PeekYou did exactly that:
Whereas Gingrich rates 8% real followers, Sarah Palin is closest with a 20% ratio of real followers, by the firm’s analysis. Mitt Romney has 26%, Michele Bachmann 28%, and Tim Pawlenty 32%. In other words, Gingrich has by far a higher proportion of fake accounts following him than any of his competitors.
Now that IS interesting. But how does one explain the large discrepancy between PeekYou’s findings and ours and The Center for Complex Networks and Systems Research’s? Gawker again:
“Keep in mind that the PeekYou weeded out business and brand accounts as not real people, which CCNS didn’t and would change the ratios.”
RAAK is not convinced I have to say.
So what CAN you take away from this article!? Newt has a stranger than most Twitter follower composition. And spotting fake followers en masse is about as easy as spotting a fake orgasm. You can play around with the data yourself here (Google Docs).
Posted by Wessel van Rensburg