Monday, March 10, 2014

The biggest Oscar winner: the data scientists!

For the past two seasons of the Oscars, a group of data enthusiasts have been blogging predictions (they also cover sports and political elections) for the winners of the big night, and they’ve been doing it rather accurately. In 2013, they correctly predicted 19 out of 24 categories, and only three of those were truly major upsets given the margins of error provided. For this year’s Oscars, they hit 21 of 24 again with only three major upsets.
Who were you thinking would win (or should have won) best actress, for example?



I hadn’t even heard of the film Blue Jasmine prior to reviewing the predictions and no one I knew had been talking about it. But everyone I knew had seen Gravity and were talking about it. If water-cooler predictions were anything to put money on, I would have guessed Sandra Bullock would be winning an Oscar this year. Fortunately, I had the folks at PredictWise to help me out. When it came time to throw in my vote for the likely Best Actress winner, I knew the smart money was on Cate Blanchett.

As Oscar night unfolded, David Rothschild kept his PredictWise blog up to date on the accuracies of his team’s predictions—as well as his choice of beverage for the night (beer, always a winner for me). The final predictions for the Oscars were posted on March 1, and as the night progressed, it seemed as if 2014 might not be going the data scientists’ way—an hour into the show, only six out of nine categories were accurately predicted! But the rest of the ever-lengthy proceedings would go their way and by 9 p.m. PST, they wrapped up Oscar night with the Best Picture award and 21 out of 24 categories correctly predicted.
Not surprisingly, the most important data elements feeding the predictive formulas were the outcomes of awards shows preceding the Oscars. You can see from the chart from the 2013 Oscars below that the error rates dropped as more award shows results came in.

If you want to get into the formulas and the data behind all the predictions, you can read all about it in PredictWise’s (not yet published) academic paper on the matter here.  


No comments:

Post a Comment