Twitter Predicts Box Office Revenue

In the first paper I’ve seen using social media (like Twitter) to tie to a real world monetary outcome, Asur and Huberman (2010) at the Social Computing Labs in Palo Alto, CA use Twitter activity to predict film box office sales.  Taking all 24 major film releases between November 2009 and January 2010, the number of tweets on new films predict sales even better than the Hollywood Stock Exchange (HSX), an online prediction game (think fantasy football for film success) that is described as the “gold standard” for predicting box office revenues.

Especially amazing is just how powerful this prediction model is – the adjusted R^2 predicting first weekend box office sales from average number of tweets per hour is 0.80.  That’s amazingly high – in other words, 80% of the variability in box office sales can be explained by variability in tweets.  And that’s a relatively simple metric!

When examining this pattern over time and including the number of theaters that the movie is released in, that number jumps to 97.3%, which is slightly better than the 96.5% achieved by HSX ratings plus number of theaters.  Keep in mind that the HSX is filled with people explicitly trying to predict film success, while the Twitter method uses only observed chatter.  That’s a powerful tool.

The problem moving forward, of course, is that 1) if Twitter is used to predict film success, 2) more people will pay attention to Twitter as an indicator of whether or not to see a movie, 3) the film industry will begin even more aggressive viral/covert marketing strategies using these technologies, and 4) the value of this measurement system decreases.  But for now… it’s certainly an impressive find.  The authors comment:

At a deeper level, this work shows how social media expresses a collective wisdom which, when properly tapped, can yield an extremely powerful and accurate indicator of future outcomes.

That might be a little generous, but it does lead me to wonder – what other organizational outcomes could we predict?