By Kent R. Kroeger (January 15, 2021)
Since we are a mere 24 hours away from the start of the NFL Divisional Round playoffs, I will dispense with any long-winded explanation of how my data loving robot (Beadle) came up with her predictions for those games.
Suffice it to say, despite her Bayesian roots, Beadle is rather lazy statistician who typically eschews the rigors and challenges associated with building statistical models from scratch for the convenience of cribbing off the work of others.
Why do all that work when you can have others do it for you?
There is no better arena to award Beadle’s sluggardness than predicting NFL football games, as there are literally hundreds of statisticians, data modelers and highly-motivated gamblers who publicly share their methodologies and resultant game predictions for all to see.
Why reinvent the wheel?
With this frame-of-mind, Beadle has all season long been scanning the Web for these game predictions and quietly noting those data analysts with the best prediction track records. Oh, heck, who am I kidding? Beadle stopped doing that about four weeks into the season.
What was the point? It was obvious from the beginning that all, not most, but ALL of these prediction models use mostly the same variables and statistical modeling techniques and, voilà, come up with mostly the same predictions.
FiveThirtyEight’s prediction model predicted back in September that the Kansas City Chiefs would win this year’s Super Bowl over the New Orleans Saints. And so did about 538 other prediction models.
Why? Because they are all using the same data inputs and whatever variation in methods they employ to crunch that data (e.g., Bayesians versus Frequentists) is not different enough to substantively change model predictions.
But what if the Chiefs are that good? Shouldn’t the models reflect that reality?
And it can never be forgotten that these NFL prediction models face a highly dynamic environment where quarterbacks and other key players can get injured over the course of a season, fundamentally changing a team prospects — — a fact FiveThirtyEight’s model accounts for with respect to QBs — — and the reason for which preseason model predictions (and Vegas betting lines) need to be updated from week-to-week.
Beadle and I are not negative towards statistical prediction models. To the contrary, given the infinitely complex contexts in which they are asked to make judgments, we couldn’t be more in awe of the fact that many of them are very predictive.
Before I share Beadle’s predictions for the NFL Divisional Round, I should extend thanks to these eight analytic websites that shared their data and methodologies: teamrankings.com, ESPN’s Football Power Index, sagarin.com, masseyratings.com, thepowerrank.com, ff-winners.com, powerrankingsguru.com, and simmonsratings.com.
It is from these prediction models that Beadle aggregated their NFL team scores to generate her own game predictions.
Beadle’s Predictions for the NFL Divisional Playoffs
Without any further adieu, here is how Beadle ranks the remaining NFL playoff teams on her Average Power Index (API), which is merely each team’s standardized (z-score) after averaging the index scores for the eight prediction models:
And from those API values, Beadle makes the following game predictions (including point spreads and scores) through the Super Bowl:
No surprise: Beadle predicts the Kansas City Chiefs will win the Super Bowl in a close game with the New Orleans Saints.
But you didn’t need Beadle to tell you that. FiveThirtyEight.com made that similar prediction five months ago.
Send comments to: email@example.com