2019 NFL Postseason Predictions from Machine Learning Model — Super Bowl Update
Chiefs are favorites and trending up; how 49ers can upset
Check previous articles for more model details and context: Pre-Wild Card Round, Pre-Divisional Round, Pre-Championship Sunday
Super Bowl LIV is just a few days away! The San Francisco 49ers and the Kansas City Chiefs are in Miami, going through their final preparations before the championship game on Sunday. For the last few weeks, I’ve been tracking the performance of several prediction models that were published at the start of the playoffs. While my very own V 4.0 was leading for the first couple weeks, ESPN’s NFL Football Power Index (FPI) has taken the lead after the conference championships and clinched victory at the Super Bowl due to its higher confidence in both the 49ers and Chiefs relative to all the other models.
Now that we know who will be playing on Super Bowl Sunday, here’s a look at the updated model predictions from V4.0, ESPN’s FPI, and FiveThirtyEights’s Elo & QB-adjusted Elo.
It’s a little strange to see a number 1 seed as an underdog to a number 2 seed, but the Chiefs have been near unstoppable, and seem to be improving more than everyone else (more details in the Trending up at the Right Time section). While all of the models favor the Chiefs, V 4.0 is least confident that Andy Reid will be taking the Lombardi trophy back to Kansas City. Should we buy into V4.0’s skepticism or should we lean toward the other models, which have a more bullish view of KC’s prospects? Reviewing V4.0’s performance over the last few weeks can provide some insight and help us understand the model’s strengths and weaknesses.
Reflecting on V 4.0
V 4.0 was did well in the NFC bracket going 4-1. (The 1 incorrect prediction was an overtime win by the Vikings, which was predicted to be close: 52%/48%.) However, the model struggled in the AFC, at 1-4. It expected less from the Titans and incorrectly predicted both Texans’ games (win vs the Bills and loss vs the Chiefs). Given a prediction accuracy of 68.5%, the result of 5 or more errors had only a 17% chance of occurring.
What went wrong? Well it could just be bad luck–getting 5 games correct was about as likely as “crapping out”, which occurs infrequently, but certainly more often than shooters would like. However, as I mentioned in the Championship Sunday Update, there are some weaknesses to V 4.0 that may have led to the errors. (I already discussed in detail why V 4.0 may have been undervaluing the Titans, and while V 4.0 TA moved predictions in the right direction, it was likely an over adjustment.) What about those Texans’ games? One explanation is that V 4.0 had a hard time assessing the Texans’ defense. Houston’s defensive unit declined following JJ Watt’s injury in week 8, was bolstered just enough by his return to defeat the Bills in overtime, but was then scorched by Patrick Mahomes and the Chiefs’ offense, allowing 51 points. A look at the Overall Offense and Overall Defense metrics [numberFire], over time, which were inputs for V 4.0, can help explain some of the erroneous predictions.
Trending up at the Right Time
There is a lot of interesting data in these two charts! Let’s focus on a few trends to help us understand V 4.0 and make the best possible prediction for the Super Bowl. First, let’s focus on the Overall Offense of the Titans and the Chiefs. You can see both lines flatten/dip around the time their top quarterbacks were not playing, but then steadily increase after Mahomes returned from injury and Ryan Tannehill was named the starter. Since this metric is a running aggregate, it takes time to “catch up” to significant changes and thus, likely underestimating the true offensive prowess of these teams when their best signal callers were playing. What’s most relevant for the Super Bowl is the steepness of the increase for the Chiefs. Their offensive dominance has continued to increase at a stellar rate, suggesting that even this late in the season, they might still be underrated. The 49ers offense is also trending up, but not as sharply as the Chiefs.
When it comes to defense, there are two clear outliers, the 49ers and the Patriots. They both steadily improved during the regular season while most other teams declined or plateaued. However, late in the season their Overall Defense has remained steady and possibly showing some signs of regression, suggesting that if anything, V 4.0 might be overestimating their defense slightly. As for the Texans’ defense mentioned above, they had the most drastic negative change near the end of the season with some variability mixed in, which explains why V 4.0 had a hard time predicting their performance week to week.
What’s the upshot for Super Bowl Sunday? It seems that V 4.0 would improve in accuracy by incorporating trends–and the trends favor the Chiefs. In other words, V 4.0’s prediction of 56% for a Chiefs’ victory is likely an understatement and it wouldn’t be a total shock if they happen to win comfortably as they did against their previous postseason foes.
How the 49ers might pull off the upset
We’ve seen the Titans defy the odds as underdogs in the playoffs–twice. Kyle Shanahan and the 49ers can do it too, but they’ll likely need a few things to break their way in order to win.
In their two upsets, the Titans won the turnover battle, limited their opponents to more field goals than touchdowns, and made some big short yardage stops on defense. They were not able to pull off any of those feats in their loss to the Chiefs. The 49ers will likely have to accomplish those tasks in order take Lombardi Trophy back to the Bay Area. They might even need to pull off a trick play or two like Derrick Henry’s jump pass. The other possibility we cannot ignore given the nature of football, is that an injury to a key Chiefs’ player can help SF’s chances, but let’s hope it doesn’t come to that.
Heart vs. Head
I grew up in Chicago and my first allegiance is the Bears. However, I’ve been living San Francisco for the last 3 years and watched John Lynch & Kyle Shanahan turn the franchise around from 4-12 last year to 13-3 and a Super Bowl appearance. I’ve quickly become a fan, and since the Bears are out, I’ll be cheering for my new hometown team even though my computations say a 49ers victory is unlikely. If they can get a key turnover or two, somehow make a few short yardage stops against a team with an abundance of offensive weapons, and/or surprise the Chiefs with some well-executed trick plays, then we just might be celebrating in San Francisco.
