Best Picture Prediction: Experts vs the Machine

This was going to be a contest of Experts vs. the Machine.  But I never heard back from the experts I contacted.  Except Ben Atwater.  Ben is the kid who runs the Collective.  He is also our in-house movie critic. You can see his latest reviews: Lego Batman, Godzilla, and Batman v. Superman (and there are many, many more reviews by Ben on the Collective).  Ben knows his stuff. In fact, it was Ben came up with the idea of a having a Machine vs. Experts contest.

I will represent the machine.   I had to rely heavily on Ben for domain knowledge – my knowledge of the movie business and the mechanics and inner-workings of the academy awards is skimpy.  Ben helped put together the data set on which the machine prediction is based.

 Here is Ben’s prediction for Best Picture:



Best Picture Prediction

Ben Atwater

La La Land


And here are the predictions of the Machine.  I ran two models and both gave up the same result. I present the results graphically.


 The first model, a logistic regression, predicts Moonlight in a three-way horse race with La La Land and Hacksaw Ridge.  Admittedly, the differences among and between the three are not statistically significant.  But I will stick with Moonlight.

 The second result come from a Naïve Bayes classifier; and it predicts Moonlight as well, nipped at the heels by La La Land.

 So there you have it.  The Machine predicts Moonlight; the Expert unanimously pick La La Land.



Best Picture Prediction

Ben Atwater

La La Land

The Machine



What are others predicting?  Fivethirtyeight has been consistently picking La La Land; here they are on February 6 “it’s become increasingly clear that “La La Land” will likely win best picture.” Here they are on February 20 – La La Land still strong.

 What about the wisdom of the crowd? Again, La La Land.  Oddshark for instance – on 2.24.17 - states “La La Land the odds-on favorite to win Best Picture.”  Over in the UK, NicerOdds – on 2.24.17 - has La La Land at 1.12 to 1 and Moonlight at 5 to 1.  This is equivalent to a 90 percent chance of winning for La La Land and a 20 percent chance of winning for Moonlight.  The Machine has Moonlight at 19 percent.  So we agree on something.

The final tally of all those surveyed for this event is as follows:



Best Picture Prediction

Ben Atwater

La La Land


La La Land


La La Land

The Machine



How did the Machine predict? We used the logarithmic regression model in the Base R package and the Naïve Bayes model in the e1071 package.

The variables we used and the description is the following:



Run Time

There is a perception that deserving winners have to be long.


Whether the movie was based on a book

Golden Globe Nomination

Whether the movie was nominated for a Golden Globe award

Golden Globe Winner

Whether the movie won a Golden Globe

Party in Power

Democrat or Republican


Dow Jones in Levels 1-month prior to the awards ceremony date

Dow-Jones change

Change in the Dow Jones relative to previous year; calculated 1-month prior to the awards ceremony date

Social Relevance

A 1:7 Likert Scale where the highest social relevance is a 7

Opening Gross

Gross Revenues on the Film’s Opening Day

Number of Theaters

The number of opening day theaters.

Regime dummy

Prior to 2009 there were at most 5 films nominated; after 2009 there were up to 10.


Most of the movie-related data came from Box Office Mojo and The Numbers; Dow Jones data and the deflator came from FRED. Social Relevance and Originality were put together by Ben Atwater.  The nature of the variables suggests that the machine is implicitly assuming that the best picture decision reflects the “times” the “zeitgeist.”




Email me when people comment –

You need to be a member of UNH Economics Collective to add comments!

Join UNH Economics Collective