A tweet issued as overtime started in the Super Bowl that traced the win probability graph for the New England Patriots and Atlanta Falcons from the beginning of the game showed a startling similarity to a chart that could have traced the chances of winning the presidency for Donald Trump and Hillary Clinton on Election Night 2016.
Here’s the graph from the Super Bowl:
After the Falcons scored a touchdown to stretch their lead to 28-3, the graph showed the Falcons’ chances of winning at nearly 100%. Even after the Patriots scored a touchdown and a field goal to cut the lead to 28-12, the graph showed the Falcons’ chances of winning over 90%, as time seemed to be running out. The Patriots scored another touchdown, trimming the Falcons’ lead to 28-20, but the graph, noting the paucity of time left to the Patriots, still hovered over 90% for the Falcons.
After the Patriots scored another touchdown to tie the game, the Falcons were still favored, as just enough time remained for them to mount a drive for a game-winning field goal, which did not eventuate.
Similar prognostications of doom for Donald Trump were made by virtually every pollster in the 2016 presidential election. As a classic example, let’s take a quick look at how Nate Silver’s 538.com, which has been historically incredibly accurate, saw the chances of Donald Trump change radically on Election Night; all times are Eastern Standard Time:
5:49 p.m. “She now has an 85 percent chance of winning according to an aggregation of two betting markets, Betfair and PredictIt.”
6:53 p.m. “How important is Florida? (Polls in the eastern part of the state close in a few minutes.) If Clinton wins it, her probability of winning the Electoral College would shoot up to 93 percent from 71 percent, according to our election night model. And if Trump were to win it, his chances would increase to 59 percent from 29 percent.”
7:05 p.m. “Clinton wins Vermont. Our model now gives her a 72 percent chance of winning the election.”
8:03 p.m. “Clinton wins Delaware, Illinois, and Rhode Island. Our model now gives her a 78 percent chance of winning the election.”
8:58 p.m.: Nate Silver warned, “We’re in something of a state of suspended animation right now. The results so far are pretty well in line with pre-election polls, which showed a close race in many swing states and Clinton more often having the lead. But the problem with having a lot of narrow leads is that you don’t always convert them to wins, and so far no major states have fallen to Clinton.”
9:08 p.m. Silver warned again, “The Upshot’s calculator has Clinton projected to eventually win the popular vote by 3 to 4 percentage points — right in line with where national polls had the race. And yet, they show her winning only 290 or so electoral votes, which obviously means that Trump has a decent shot to win the Electoral College.”
10:12 p.m., even before Florida was called for Trump: “Clinton wins New Mexico. Our model now gives her a 60 percent chance of winning the election.”
10:24 p.m. “The presidential races in Wisconsin and Pennsylvania are too close to call. Our model now gives Clinton a 51 percent chance of winning overall; Trump now has a 48 percent chance.”
10:25 p.m. “Trump wins Ohio. Our model now gives him a 55 percent chance of winning the election.”
What is interesting, in retrospect, is to notice that none of the big states that eventually yielded the win for Trump had been called yet; the jury was still out on Florida, Michigan, Wisconsin, North Carolina, and Pennsylvania. But 538 was seeing the trend in the states drift toward Trump, so even before Trump’s victory was certain, they started shifting the chances of victory for Trump.
The Super Bowl graph was an accurate reflection of the Patriots’ chances, as were the predictions of pollsters for election night, given the information available at the time. But the narrow victory of the Patriots against tremendous odds, and the victory of Donald Trump against what seemed like insurmountable odds are a telling reminder that nothing in life is certain until it’s over.
