The Win Probability Changes From Super Bowl LI Were Mind-Blowing

The Patriots won Super Bowl LI in the most unlikely fashion imaginable.

Last night, I learned that Tom Brady isn't made from the same stuff you and I are made from.

I also learned that people love to hate on win probability models.

For the uninformed, a win probability model's goal is to provide odds of a team winning or losing a game at any point during that game. To figure that out, an algorithm is created to calculate the win probability added (or lost) on each and every play.

So, you can imagine that the biggest win probability swings occur when, say, a pick-six happens. Or late in a game when a prayer of a Hail Mary hits. Or when there's a punt return for the win.

Or when a team comes back from a 25-point deficit in the second half.

Super Bowl LI Was a Sincere Outlier

When the Score Was 28-3
If you watched the Super Bowl, the term "blowout" crossed your mind at least once. The Falcons were sitting so, so pretty in the third quarter after scoring a touchdown to make it 28-3. In fact, at that point in time, numberFire Live, our live win probability platform, gave the Falcons a 97.62% chance to win the game.

That's too high, you say? A lot of people said that.

But over the last 20 years, only two teams have actually come back to win a game after being down 25 or more points at any point in the third quarter. You might remember one of them: it was the big Andrew Luck breakout playoff game, where he brought his Colts back from their deep deficit against the Chiefs in 2014's wild card round. You definitely remember the other one: it happened last night.

It's not as though a team doesn't often trail by this margin in the third quarter, either. During this 20-year timespan, 433 NFL games have featured a team trailing by 25 or more points in the third quarter. That means 0.46% of the teams who've been down by that much actually mount a comeback.

Maybe 97.62% was conservative. (Or maybe it was accounting for the fact that Tom Brady was quarterbacking the team that was trailing.)

When the Score Was 28-12
That win probability actually rose during the game, too. After a red zone stop and a Stephen Gostkowski field goal in the fourth quarter, the Falcons had the ball on their own 35-yard line. It was 2nd-and-2. At that point in time, Atlanta's probability to win was at 98.66%. Tevin Coleman then ran the ball for one yard, and on 3rd-and-1, Matt Ryan fumbled the rock, giving the Patriots amazing field position.

When the Score Was 28-20
After the Patriots scored and saw a successful two-point conversion -- this was with just under six minutes to go in the game -- the Falcons still had an eight-point lead with the ball. Devonta Freeman then converted a 39-yard reception, and then Ryan completed a pass to Julio Jones, who made a circus catch on the sideline.

So let's bring this in for a second. After the what-could-have-been-a-historically-famous catch, the Falcons had a first down on the New England 22-yard line with 4:40 left in the game. And they were up eight points.

Their win probability at this time was 96.55%.

I'll use the last 20 years for reference again: Since 1997, only three teams have lost a game after being up by eight or more points within an opponents' 25-yard line with less than five minutes to go. And it hasn't happened since 1997.

The most important set of events occurred after that Jones catch, though, which allowed the Patriots to come back and win the game. Freeman was held for a one-yard loss, and then Ryan took a 12-yard sack. Still in field goal range, the Falcons then took a holding penalty, which forced them to punt.

The thing is, Atlanta still had great odds to win the game even after the botched drive. Their punt gave the Patriots field position on their own 9-yard line, meaning New England had to go 91 yards to score a touchdown to try and tie things up with time restrictions. And, of course, the Pats still needed the two-point conversion.

At the start of the Patriots' final drive in regulation, they had just a 7.66% chance to win. It was far from a given that they'd come back and win the game.

But they did, making history.


Parting Thoughts

I understand why Patriot fans might give the middle finger to a win probability model that showed the Falcons as near locks to win Super Bowl LI late in the game. But what needs to be understood here is that probability is just that -- it's probability.

New England beat the odds over and over again in this game. They did something we've never seen before.

The truth is, any win probability model that gave the Falcons an incredibly high chance to win Super Bowl LI in the second half was actually doing its job. It was actually correct based on everything that's ever happened in the history of the NFL.

What a model like this does is contextualize the incredible feat New England just pulled off. Their comeback was that unlikely. Their play was that good.

Their Super Bowl win was that amazing.