The Oscars are predictable. And there is a science behind such predictions.
While the degree of accuracy varies, from one predictive model to another, one nominee stands out in each category, every year. For instance, it can be predicted with 99.8% certainty that, this year, Joaquin Phoenix is going to win Best Actor in a Leading Role for Joker . Whereas in the Best Picture category, the win probabilities are relatively more closer, with 1917 at 71%, Joker at 10% and Parasite only at 2%.
But how do statisticians predict the Oscar winners? More importantly, how many past winners coincide with the predicted?
While the mathematical model involves complicated equations, the basic premise is simple.
Ever played the game of battleship? While the odds of spotting an enemy ship are initially slim, the probability increases as the game goes on. The Oscar predictions follow the same principle: when an exercise repeats itself over a period, it creates enough patterns to guess the future outcomes. With the 92nd Academy Awards to be presented on February 9, mathematicians have 91 previous results as data to build a model using which the future winner can be predicted.
Iain Pardoe, who has a doctorate in Statistics from the University of Minnesota, has built one such model, which has proved to be one of the most accurate.
He uses three basic patterns.
Firstly, the Academy members involved in deciding the Oscars are also present in the panel which decides the Golden Globes, Directors Guild of America, Screen Actors Guild, BAFTA etc. Thus, their choices can’t wildly vary between the two awards. For instance, of the 64 Best Picture winners from 1943 to 2006, 34 had previously won the Golden Globes. Thus, the awards preceding the Oscars is an important data point.
Secondly, winners of the Best Picture and Best Director are often represented by multiple nominees in other categories. For instance, between 1938 and 2006, only three movies have won the Best Picture without also receiving a Best Director nomination. Thus, the inter-relationship between nominations adds to the final probability.
Thirdly, past nominations of the lead actors and directors increases their chance of winning. Whereas, past wins decreases the chance of a repeat in case of Lead Actor and Lead Actress.
Weights are then assigned based on which of the above assumptions correlates strongly to the past Oscar winners. Using that relationship, the 2019 winners are predicted. The following table shows the predicted winners in the four major categories — Best Picture, Best Actor in a Leading Role, Best Actress in a Leading Role, Best Director — and their win probabilities.
The winners are predicted using our data-driven model
A word of caution though. No model can be 100% accurate. Between 1938-2018, the chosen model has accurately predicted 71% of the Oscars winners.
The odds become better with a 80% success rate, when only the last 14 years are considered.
However, the Academy has thrown up surprises on many occasions. Especially the Best Picture category — as it represents a collective effort and not an individual's brilliance — have often gone off the charts.
Notable exceptions in recent years: 2016 winner Moonlight (had only 2% probability of winning, but beat La La Land which had a 97% chance); 2015 winner Spotlight (with just 6% probability beat The Revenant which had 56% chance).
This time too, there could be exceptions with the Korean movie Parasite riding the race as a dark horse.
