Bing Predicts Calls the Super Bowl, Caps Off Great Season

On Sunday night, the New England Patriots hoisted the Lombardi Trophy for the fourth time this century, just like Bing Predicts said they would.

bing predicts super bowl

This was no bandwagon pick by the search engine. Bing has been touting the Patriots as the likely winner since early December, long before it was even clear which teams would be in the playoffs. The Patriots Super Bowl win gave Bing a 9-2 record for the playoffs and a 178-89 record overall for the entire season.

Okay, so this was hardly a bold or unprecedented prediction, but everything about Sunday night's game emphasized the success of Bing's predictive model.

Bing Predicts Calls the Super Bowl

Heading into Super Bowl XLIX, Bing declared a Patriots victory in what the search engine promised would be an exciting game. According to Bing, the chance of a Patriots victory was 51.5%, and the final score would be 24-23. In other words, this was going to be anyone's game.

odds of seattle winning

It did indeed come down to the very end. It was indeed anyone's game. We could spend all the time we want discussing the ridiculousness of Seattle's final play call (give the ball to the Beast!), but such discussion wouldn't change the fact that Bing nailed it.

Bing may have missed the final score by a few points, but the search engine's declaration of a 51.5% chance of a Patriots victory was much more accurate than the Vegas spread. This was a game where two outstanding teams had an almost equal chance of winning. Maybe Bing was looking into the same crystal ball the Patriots must have used to defend against the pass on that final Seattle play.

There was no favoritism here. Bing called this game fair and square, not showing any allegiance to Seattle Seahawks owner and Microsoft co-founder Paul G. Allen. Was Mr. Allen disappointed that Microsoft's search engine didn't believe in his team? Maybe not, but it does make for a somewhat interesting twist.

Yes, a Search Engine Can Predict the Outcome of Football Games

There were plenty of naysayers all season who thought Bing should have done better. Many a keyboard hero posted a blog comment about how Bing's record was unimpressive. A supercomputer should be able to do better. I can predict games better without an algorithm. Big deal; Bing's not even picking against the spread.

Here's the thing though: Bing wasn't just trying to pick the winner. Bing was trying to create an algorithm that predicted the likelihood a team had to win. While that may seem the same to some people, it isn't at all. Saying the New England Patriots have a 51.5% chance to win is not the same as just saying the New England Patriots will win.

So what's the difference? And where's the proof that Bing was successful?

Well, simply saying a team will win does not necessarily rely on statistics. Nor can it be proven to be anything more than a lucky guess. Predicting the likelihood of a particular outcome, on the other hand, is based purely on data, not on gut feelings, and we actually can find some proof in the pudding.

The Case for Bing Predicts

Of the 267 football games played this season, Bing correctly picked the winner 66.7% of the time, which more than makes good on the search engine's promise to get most of the games correct. However, 66.7% is hardly an impressive record. There are plenty of other predictive models that did just as well (or better). That's not the point. Bing sought out to create something more than another betting tool to tell you who would win.

Bing Predicts is a much more sophisticated experiment, and not just because it uses a combination of historical data and social signals.

So what makes Bing so special?

First off, it's important to understand that it's impossible to prove that Bing was ever wrong. The outcome of an individual football game, regardless of the score, can never tell us the actual chances each team had to win. Had the Seahawks won by 43 points on Sunday night, we couldn't definitively say that the Patriots didn't have a 51.5% chance to win. Now, if the two teams played 10 times and Seattle won all of them, then we would seriously doubt Bing's accuracy. (I have a feeling if those two teams played ten times, we'd probably see 5 wins from each.)

While that's all nice, is there any proof that Bing was actually correct with its predictions? As a matter of fact, there is.

Something very interesting happens when we break down Bing's predictions into smaller increments:

bing predicts graph

Let's take a close look at the data here. When Bing predicts a team has a 50-59.9% chance of winning, that team wins 52.6% of the time. These games are virtually toss-ups. When the chance of one team winning is under 60% there's a pretty good chance that either team can win. What happened in these games? Bing had the right team winning a little over half the time. Looks pretty good so far. But it gets better.

If a team has a 60-69.9% chance to win, things are looking a bit more certain. In these games, Bing calls the winner 71.43% of the time. When the chances go up to 70-79.9%, Bing's accuracy goes up to 75.3%. When the chances rise to 80% or above, Bing is right 79% of the time.

It's obvious what's happening here. Bing is more accurate in predicting the winner when it predicts higher chances. This should be true, but it need not correlate exactly in order to be proof. After all, if I flip a coin 100 times and it lands on heads 59 times, that doesn't necessarily mean the odds aren't 50-50.

There are a few more pieces of data that help prove Bing's prediction success. When Bing sided with home field advantage, the home team won 67.6% of the time. While that may not sound impressive, the home team historically only wins about 57% of the time. When Bing picked the visitors, it was correct 64% of the time. Both of these numbers are well above the historical trends.

Perhaps the most impressive stat of all for Bing Predicts is this: if you average Bing's final predicted chances for each game throughout the season, you end up with 66.0%, which is mighty close to Bing's overall accuracy of 66.7%. Conclusion: Bing knows football.

What's Next for Bing Predicts

Don't expect Bing Predicts to go away until next football season. This is a prediction engine that focuses on a lot more than sports. Bing Predicts looks at fashion, politics, awards shows, technology and much more. It is both a statistical algorithm and an analysis of our daily lives. Heck, the search engine even has insight into what we're going to eat and where we're going to go on vacation.

We shouldn't expect Bing to begin predicting everything about the future, but we should see the search engine continue to expand. Although Bing trails fair behind Google in the search game, it's doing amazing things with data.

What would I like to see next from Bing? I'd like the search engine to release a completed March Madness bracket before the first tip-off. I'd like to see it predict weekly box office numbers. I'd like to see it do so many other things. Not creepy things though. I don't want Bing to predict my life span based on my search history. Nor do I want to see Bing usher us into some dystopian future where we just have to conduct a simple search to know what's going to happen in any facet of our lives.  But I don't see much chance of Bing using this technology for evil in the near future. For now, it's mostly just fun and games.

First Scribe is now Perrill Why the change?