A way to predict winners at the Oscars

[This short piece appeared in New Scientist 3/29/2008, Vol. 197, Issue 2649.]

A statistical model picks out best picture, director and actors at the Academy awards with surprising accuracy

AT LAST, relief from the tortuous uncertainty of Hollywood's annual Academy Awards. Now there is a surprisingly accurate way of predicting who will win.

Various attempts have been made to predict the Oscars, such as. Instead, Iain Pardoe of the University of Oregon in Eugene and Dean Simonton of the University of California, Davis, took a purely statistical approach. They analysed the histories of around 1600 Oscar nominees between 1928 and 2006 in the four major categories: best picture, director, leading actor and leading actress. Then the pair teased out several factors that correlated with Oscar wins, including previous nominations and Golden Globe wins.

Plugging the patterns into a statistical model, Pardoe "predicted" the winners from their track records. The model's accuracy for the 30 years leading up to 2006 was at least 70 per cent for all four categories, reaching 93 per cent for best director (Journal of the Royal Statistical Society, vol 171, p 375). "It is quite a dramatic improvement on just pure random guesses," says Pardoe.

The model scored three out of four in Los Angeles last month, predicting that Daniel Day-Lewis (below) would win best leading actor, and the Coen brothers would scoop best director and picture for their film No Country for Old Men.


Return to main page