We’ve the reached the end of the most exhausting, frustrating, and vexing awards season I’ve seen in recent memory. Throughout it all, I’ve debated on whether or not I should even bring back Statsgasm’s prediction models to Awards Daily for a sixth year. But alas, after a few late-nighters this week I’m here to announce that Statsgasm has returned!
For those of you new to the site, I am a stats guy by training, with degrees in math and economics. Statsgasm began in 2013 and they are a set of models built using a technique called regression to predict winners in 21 categories — specifically, I use something called small sample (penalized) logistic regression to estimate the chances of winning for all nominees. Regression analysis is the underlying methodology for the Oscar prediction models used by Ben Zauzmer at The Hollywood Reporter, FiveThirtyEight.com, and a few other masochistic souls out there. Although all of us will be using the same fundamental techniques and data when creating these models and their estimates, there will still be differences in our results based on variable selection, ad-hoc adjustments, and other factors. In other words, there’s some art to go along with the science behind this statistical wizardry. In my opinion, you certainly need to have a healthy appreciation of awards history to be remotely successful in this endeavor.
(If your eyes haven’t glazed over yet by this wonkiness on an film/Oscar blog and are instead interested in taking a deeper dive into what I’m talking about, please review my earlier writings on the subject here.)
As always, I will offer the following caveat before I reveal all 21 of Statsgasm’s predictions: probability by nature is not really about absolutes, so I never expect Statsgasm to go a perfect 21/21, and neither should you. In terms of probability, only 90% and above should really be considered a “lock,” and there are really only a few categories meeting that standard this year — the rest all occupy a continuum from “total free for all where everything has below a 30% chance of winning,” to “there’s a solid frontunner but it’s not a lock,” and everything in between. Statistical prediction models are dependent on both the quality of the data being used as well as the skill and expertise of the model builder. But data alone from precursors can only take you so far in predicting the Oscars well (and winning your Oscar pool).
That being said, although there are a number of ways to evaluate overall performance, Statsgasm’s models has been fairly consistent in correctly predicting the most likely winners in at least 15/16 of the 21 categories each year, and did go 17 of 21 overall last year.
Statsgasm undoubtedly faces its biggest test this season. Here are its predictions for the 91st Oscars, with commentary:
(NOTE: The following Tableau dashboard detailing Statsgasm’s prediction results is best viewed on a desktop/laptop computer)
Please feel free to ask any questions you may have in comments, or reach out to me directly! And happy predicting! Be sure to get your predictions in for AD’s contest if you haven’t done so yet!