Hot Reads: We Should Probably be Better with Improbability
Photo Credit: Aaron Babcock

Hot Reads: We Should Probably be Better with Improbability

February 06, 2017

The graphs started showing up shortly before halftime of Super Bowl LI — the win-probability graphs. Alongside Crying Jordans, bad takes and new gifs, these graphs are a product of watching major sporting events unfold on Twitter. Their purpose at the half was to tell us just how screwed the Patriots were when they went to the break down 21-3 to the Falcons.

I’m assuming you watched the game, so I’ll keep the second-half recap short: Stuff happened, Atlanta didn’t run the ball when it should have, Tom Brady kept playing New England’s brand of backyard football, two two-point conversions, overtime, Patriots win again.

The win-probability graphs came out again, this time showing New England’s unbelievable rally. According to, it was the most improbable comeback in Super Bowl history:

But graphs and numbers like that make some people cranky. At least that’s what I learned last night.

That’s from Yahoo! Sports’ Dan Wetzel, one of the best sports writer in the business, which is why I was sort of shocked to see it from him. That tweet entered my feed because it was retweeted by another sports writer,’s Dennis Dodd. It was a popular sentiment on Twitter last night.

I’m trying to understand why these charts chafed so many here, but it’s not coming to me. What, exactly, is so maddening about a statistical model? And to be clear, it is just a model based on thousands of games played before it. It’s not predictive, but probabilistic. When Atlanta took a 28-3 lead in the third quarter, according to FiveThirtyEight’s model, just 0.4 percent of teams (over whatever span it used to build that model) had come back to win, odds you could realistically apply to the Patriots.

Wetzel’s argument is that this is all dumb because the model doesn’t know that Brady is the GOAT and plays for the Patriots. At least that’s the first part of the argument. Say you did factor in Brady’s greatness, how much does it change the probability? Maybe a couple tenths of a point? It would be a terrible model if it spit back “the average team in New England’s spot loses this game 99.6 percent of the time, but with Brady at quarterback the Patriots are actually favored while being down 28-3 with 23 minutes of clock left.” And if that’s not the complaint, then we’re arguing over fractions of a percentage point. The model didn’t say Atlanta would win, only that it was extremely likely. 

As for the play-calling argument, you could argue that Atlanta offensive coordinator Kyle Shanahan would’ve been better off if he had his own real-time look at the probabilities. Maybe then he would’ve realized that the tradeoff between a touchdown and a field goal wasn’t worth the risk, called (run) plays designed to ensure the Falcons got a field goal instead of the high-risk passes that started the snowball down the mountain and we wouldn’t be here at all.

What would have happened then? What if win probability had helped Atlanta win the Super Bowl? Still stupid?

For a certain faction, they’re just never going to like this approach even if it helps with the myth-making columns. I guess an unwillingness to think probabilistically is good for job security, but win-probability graphs are a thing now any time there’s a massive comeback. Dismissing them seems like the wrong side of history here.

(OK, rant over. I suppose we should at least make some attempt to link Nebraska in here, so congrats Vincent Valentine on the Super Bowl ring. His rookie season certainly wasn’t what you’d call probable when he declared for the draft. I didn’t see any graphs on that, but if I had I know they would have at least left open the possibility of what actually transpired. That’s just how probability works.)

The Grab Bag

Today’s Song of Today

  • Never miss the latest news from Hail Varsity!

    Join our free email list by signing up below.

Share via
Copy link
Powered by Social Snap