Oscars predictable as ever; in fact moreso...
Sun, 30th Mar 2008
Anyone who fancies a punt on the winners of next year's Oscars may be able to boost the success of their predictions with the help of statistics.
US-based researchers Iain Pardoe and Dean Simonton studied the results of 1600 Oscar-nominees from 1928 to 2006 for best picture, director, leading actor and leading actress. The duo were then able to dissect out several predictive Oscar-winning ingredients such as previous nominations and Golden Globe wins. By feeding the patterns into a statistical model the researchers were able to correctly identify "winners" each year just from their track records. From 1976-2006 the model, which the researchers have published in the latest edition of the Journal of the Royal Statistical Society, hit the nail on the head with 70% accuracy; for best director it was right 93% of the time.
According to Pardoe, "it's quite a dramatic improvement on just pure random guesses." In the current round of awards the model maintained its track record and correctly identified three out of four winners last month in LA including Daniel Day-Lewis as best leading actor and the Coen brothers as best director for their "No Country for Old Men".