We often overestimate the likelihood of success of viral hits.
In 1948 Harold Lasswell defined the objective of media communications research as discovering “who says what to whom in what channel with what effect”. The difficulty for researchers has been that for the first half-century or so after Lasswell set out the aim, it has been largely impossible.
Perhaps that is now changing. Online social networks generate a huge amount of information about who says what to whom. Economists, computer scientists and sociologists are now digging through these social networks for the answers to long-standing questions – and few answers are as eagerly awaited as the secret of producing a sure-fire hit.
So how do you produce the perfect film or write the perfect book – or compose the perfect tweet on Twitter – in a way that will maximise the chances of catching on? Duncan Watts, a mathematical sociologist at Columbia University and Yahoo! Research, has answers – and I’m afraid they’re not too encouraging. “I’ve been using social media to promote my book,” he says, “and it’s just a waste of time – it has almost no impact at all.”
(I’ll throw him a bone. His book is called Everything is Obvious (Once You Know the Answer).)
Part of the problem, perhaps, is that our expectations are skewed. If you ride on London buses, you may be astonished to discover that many of them are almost empty. The average London bus, according to the UK’s Department for Environment, Food and Rural Affairs, contains only 17 passengers. Clearly most bus-riding people are travelling on the full ones.
It’s a similar story with viral media: we notice the successes simply because they are successful, and overestimate the likelihood of success. And there’s a survivor bias: in our analysis of what works we ignore what fails. “People think it’s all about videos of cats or cute children,” says Watts, “But there are millions of videos that have these attributes but haven’t spread.”
Watts and his colleagues, in a research paper titled Everyone’s An Influencer, place numbers on a specific type of media hit called a “Twitter cascade”. A Twitter cascade occurs when one person’s “tweet” (short message) is repeated by other users (“retweeted”), whose retweets are themselves retweeted further, and so on. Anyone unfamiliar with Twitter can imagine a particularly good joke or piece of gossip spreading.
The first surprise, then, is that the typical Twitter cascade is both rare and tiny. Ninety per cent of tweets are never retweeted, and most of the remainder are retweeted only by a person’s immediate followers, not by those at two or three removes.
The second surprise is that beyond the mind-numbingly obvious, it’s impossible to predict which tweets will start cascades. Simply knowing that a user has started previous cascades tells Watts and his colleagues almost everything they can divine about the likelihood of future cascades – which is not very much. (It is not especially useful to know how many followers a user has if you know about their previous success in starting cascades, because the two pieces of data overlap.)
Duncan Watts would like to see marketing companies running properly controlled experiments to see which messages carry through social networks such as Facebook. But he’s not convinced that they will. “When you do the experiment properly, all the numbers go down,” he says. Watts believes that the likely outcome of such experiments would be to demonstrate how difficult it is for social marketing to have any impact.
I can now barely summon the will to beg you to follow me on Twitter (username: @timharford). But, there it is. Most things fail, and as Watts says, “the curse of being able to measure everything is that you get slapped in the face with this reality all the time.”
Also published at ft.com.