By Erin Richey
How did your Academy Awards pool turn out this year? Did you accurately predict Jared Leto for Best Supporting Actor and Cate Blanchett for Best Actress? Chances are if you laid your bets based on data from past award ceremonies, social media and prediction markets, your picks won.
Big data and predictive analytics are familiar tools in marketing and product development, but they’re also popping up to predict everything from sporting events to political elections, box office sales figures and the next Pope.
For this year’s Academy Awards, Columbus, Ohio-based Farsite Group made correct predictions for all six of the major categories. PredictWise, a research project of Microsoft Research economist David Rothschild, made predictions on 24 categories and succeeded in all but three of them. And Ben Zauzmer, who writes “OscarLytics” for The Hollywood Reporter, correctly predicted 16 of 20 categories using a more limited data model.
By comparison, movie industry columnists didn’t fare as well, rarely making accurate predictions on all six major categories. For big data evangelists, the success of predictive analytics over the experts’ predictions is both a proof of concept and a way to show potential clients how useful big data can be in more conventional contexts, such as predicting customer behavior.
According to Farsite executive director Michael Gold, predicting the Oscars started as a way to get people talking about big data. “The challenges that we run into are that [big data is] a pretty esoteric topic that not a ton of people pay attention to and we’re often working with clients’ sensitive and confidential data and processes so we can’t talk about [specific examples],” says Gold. So we took the opportunity to look at the intersection of business, pop culture and big data.”
Predictive analysis for the entertainment industry relies on objective data about human reactions to a particular movie. Most models draw heavily from data about past accolades won by those involved in each film and the awards the film won in the run-up to the Academy Awards.
Previous awards “are among our most important predictors because you have, in large part, the same voters in the guild awards as those that will be voting at the Academy Awards,” says Gold.
He adds that this year other awards didn’t cause any big changes in Farsite’s predictions, but last year, “Lincoln” started as the heavy favorite for Best Picture at nomination and “Argo” — the winner — ended up being the favorite on the night of the Oscars because of the awards it had won over “Lincoln” in the meantime.
Underscoring the importance of real-time data, Rothschild says even Oscar wins early in the ceremony changed the odds for those in the running later on. In his model, the odds for “12 Years a Slave” to win Best Picture increased by seven percentage points between the start of the event and the moment it won because of its Best Adapted Screenplay win and Lupita Nyong’o’s Best Supporting Actress win.
For their Oscar predictions this year, Farsite and PredictWise also included data from the “fire hoses” of social media and prediction markets, websites where users can place bets on the likelihood of world events, politics, sports and entertainment.
Gold says that the multi-layered approach to data collection and analysis resembles what Farsite does for its clients. “What we were trying to do was understand the behaviors of a group of people — in this case, it was six thousand or so Oscar voters,” says Gold. “So in that sense we’re certainly hoping retailers and others try to understand the behaviors of hundreds of thousands, if not billions, of their customers in the drive toward some strategic objective.”
Predicting how people will behave by using computing models can be tricky. In 2013, PredictWise published its predictions for who was to replace Pope Emeritus Benedict XVI. Then Cardinal Jorge Bergoglio, appointed to the position as Pope Francis almost a year ago, was given a 2.6 percent probability of winning the papal election, behind an Italian archbishop with a 23.5 percent probability.
“I hate the word ‘fail,’” says Rothschild about that prediction. “In my mind, everything with an answer is predictable; it’s just that sometimes, the answer is almost random.”
In many cases, predictive analysis has proven to be a major benefit for product development and marketing, and data about past cinematic successes and failures is also being used in moviemaking. Rothschild says that at Microsoft he has helped predict the success of proposed scripts, which he describes as “harder, but more meaningful as far as future allocation of resources” than Oscar predictions.
Outside the world of customer prediction models, analytics companies also offer to help sports teams predict player success and failure, and media outlets are also looking for data-supported political predictions, in part due to the popularity of Nate Silver’s FiveThirtyEight blog and his predictions in the 2012 presidential election.
The only safe prediction for next year’s Academy Awards now is that the predictive models will be more accurate. “Every additional year of history is good, and we’ve improved our model, not just in our results, but in testing,” says Gold. “So, having another year of history in our model’s performance is really helpful, and that also links to how we do this in the real world.”
Erin Richey is a freelance data journalist with B2B experience.