The Broncos are up 35-28 with 1:09 left, but that doesn’t tell you much. The Broncos are up 35-28 with 1:09 left and the Ravens have the ball with no timeouts on their own 23. Still some room for improvement.
The Broncos have a win probability of 98%. Now that, there: that is something.
Sports live in the improbable. Sometimes, the improbable happens when an individual with unimaginable athletic ability commits a series of movements that results in something like perfection: think of Peyton Manning at the end of the first quarter against the Ravens, throwing an oblong projectile to the one-and-only location Brandon Stokley can catch said pigskin. Think of Colin Kaepernick—a man who can throw a football faster than most people can throw a baseball—outrunning the entire Green Bay secondary. Think of Adrian Peterson as a human being, and not as the hybrid unicorn/man-bear he revealed himself to be this season.
Sometimes the improbable is the absence of those perfections, such as the case of Rahim Moore, waving desperately at Joe Flacco’s pass in an ill-timed leap after severely misjudging the path of the ball. Despite Dan Dierdorf’s I-have-to-use-the-word-analysis-despite-its-obvious-inapplicability-here analysis, Moore was in decent position, backpedaling as if he was ready to make a lazy interception. But then, for reasons unknown, he jumps, despite the ball still being a fair distance away. Moore erred in a very human way, and for a human to make an error is perhaps the likeliest event of all, on or off a football field. Of course, the actual result of the play was quite improbable; a 70-yard touchdown with 30 seconds left to send into overtime a game that Denver would have statistically won 98 times out of 100 in regulation.
The Broncos have a win probability of 54%.
In the unlikely event you watched the Ravens-Broncos game with Advanced NFL Stats' live win probability graph showing simultaneously—which takes historical data of the exact same score, possession, down, distance and time remaining to determine how likely each team is to win the game—you knew before the Rahim Moore Hand-Waving Incident (dibs on that band name) that this was one crazy football game. Usually, win probability graphs do not look like the EKG of a heart attack patient superimposed over a Californian seismic activity chart. With the game potentially decided on each possession in overtime, the needle shakes.
The Broncos are still slight favorites, with the ball in overtime on their own 38. Manning looks hesitant to throw, the Broncos are calling more runs than many think is warranted against the Baltimore run defense. But they have the ball in overtime.
The Broncos have a win probability of 62%.
Manning uncharacteristically rolls to his right. Manning does something Sanchez-like and throws a skeet-shooting pigeon across his body, which is easily intercepted. The best quarterback in the NFL this season—and one of the best ever—has made a huge mistake.
The Broncos have a win probability of 27%.
“Ultimately, at the end of the graph, a team’s win probability will either be 1 for a win or 0 for a loss,” Brian Burke, founder of Advanced NFL Stats and proprietor of the NFL’s live win probability graphs, told me via email. “The graphs ignore misleading stats like yards, completion percentage, or fantasy-centric stats and illustrate what really matters—winning.”
For the first time since Manning’s pick-six in the first quarter, the Broncos are in danger, and this time it is much grimmer. Baltimore is on the Denver 45, needing only a field goal to advance to the AFC Championship game in one of the most improbable victories imaginable. A few Ray Rice runs gets the ball into field goal range, and the spectacle is complete.
The Broncos have a win probability of 0%.
Win probability graphs are often relatively stable. In fact, they usually look more like this:
That is to say, one team plays better than the other team because one team is better than the other team. Then the game ends, and we are all upset we didn’t drink more beer during said game. Usually, that game pictured above doesn’t have an ending that produces a graph like the one at right.
This is a visual representation of the team that we all thought was going to go home and fire their coach instead making a fierce comeback, having the game in hand, then somehow giving it away in the final seconds, restoring the sanity under which the game operated for 45 of its 60 minutes. That is to say: the Seattle-Atlanta game.
On the first drive of the second half, Russell Wilson and company marched down the field and scored in just over five minutes, ending the shutout. At this point, the score was 20-7. Due to a combination of the Falcons’ reputation, the Seahawks history of sudden offensive spurts and crushing defensive stands, it seemed only a matter of time before Seattle started playing like a mid-oughts USC squad and finished off the Falcons. That time had seemingly come. Yet, there it is, clear as day:
Atlanta has a win probability of 92%.
“The most important thing the graphs can convey is that the sport can be approached analytically,” Burke told me. “Football is insanely complex compared to other sports. There are literally billions of combinations of score, time, down, etc… But win probability is the ‘killer’ stat in any sport. It’s what an economist would call a linear utility function, which lets analysts use some relatively simple math to learn some very deep insights about the sport and its strategy. A sound win probability model is like decoding the genome of a sport.”
In practical terms, this means that win probability can be used to determine if it was a good decision for John Fox to pack it in at the end of regulation as Burke did, or to find the very pulse of the game in creating the Excitement Index and Comeback Factor. Win Probability is the killer stat because it is a storyboard, history lesson and analytical tool rolled into one bright, real-time line.
With just over 17 minutes left in the game, Matt Ryan finds a wide open Jason Snelling to put the Falcons up by 20. A quick Seattle reply brings it back to a two-score game, but barely registers on the graph. If a team is two-scores better after 47 minutes, they are usually better after 60 minutes as well.
Atlanta has a win probability of 96%.
Some might say this whole win probability business misses the point of the game. Can anyone really enjoy the brilliance of a Peyton Manning pass through a probabilistic lens? How can you embrace the emotional swings of the game if you’re grounded in a plausible reality? Moreover, who gives a shit what’s probable? This isn’t baseball, after all. The only sample size that matters, particularly in the playoffs, is a sample size of one.
A Russell Wilson touchdown with 9:19 remaining makes things interesting, a quick Atlanta punt doubly so. Now Seattle is driving, down 6 with under 3 minutes left. A deep pass to Zach Miller falls incomplete with 49 seconds left. Seattle is calm and confident sitting on Atlanta’s 27 yard line, but the win probability model says this will only work out in Seattle’s favor 1 out of every 4 times. The odds are against them. But then Marshawn Lynch receives a short pass, slips into beast mode, and rumbles down to the 3 yard line. That play alone is worth 56% in win probability. It’s first and goal on the three yard line with 44 seconds remaining. Lynch makes the touchdown a formality, and Seattle is now winning. Atlanta has 34 seconds to kick a field goal.
Atlanta has a win probability of 8%.
This criticism of win probability is to miss the point entirely. The graphs unfold in a magical sense, revealing snapshots from every game that has ever been played. It’s augmenting the reality of the contest in front of you with all of the past, everything the human memory is too flawed to incorporate. It gives you a greater appreciation of the improbable when it does happen. A 20-7 lead is not a 20-7 lead, it is a 92%-8% lead. When that 92%-8% lead becomes precisely inverted with 25 seconds left, you realize the score was never, ever 20-7. Each play in the game is a tiny addition to the puzzle, allowing us to add one more data point to the approximation of which team will win this game—which is, after all, what matters—but until the game is over, it is only an approximation. There were always 25 minutes left to play, or 25 seconds left to play, and the score with 25-whatevers left to play is but one small part of a puzzle with a dizzying number of pieces.
Several well-executed plays later, Atlanta enters field goal range, and a 49-yard field goal—which is converted about half the time—is attempted.
Atlanta has a win probability of 53%.
These graphs aren’t exciting in themselves—this is also true every other graph on the planet—because a graph is a series of lines or bars that are meaningless without context. They aren’t a disconnected story from the game they represent. They are what is likely to happen; we can only hope something entirely different will unfold, because that is what we anticipate every time we sit down to watch a game. The expected happens every single day, and we call it our lives, and we feel different things about it on a day-to-day basis, but that is not what sports are for. This is the place for something else, and the only way to know what’s unexpected is to know what is likely, and then watch the exact opposite slap you across the face and steal your nachos before you have a chance to get your pants on.
The warm-up/icing-of-the-kicker kick goes wide right, but the real attempt is as good as soul. After some squib-kick silliness and a Hail Mary that couldn’t happen with pensioned officials, the approximations are over.
Atlanta has a win probability of 100%.
Anyone can look at the final score and see who won, which most people will tell you is all that matters. But if you watched either of these games, you know that isn’t true. As Burke told me before the games kicked off, “The story is about how they got there.”