Is Twitter the best predictor of March Madness?

Southern Methodist University’s Yanick Moreira (2) drives against UCLA’s Tony Parker (23) and Bryce Alford (20) during the first half of an NCAA tournament second-round college basketball game in Louisville, Ky., March 19, 2015.
Jamie Rhodes | USA TODAY Sports | Reuters
Southern Methodist University’s Yanick Moreira (2) drives against UCLA’s Tony Parker (23) and Bryce Alford (20) during the first half of an NCAA tournament second-round college basketball game in Louisville, Ky., March 19, 2015.

March Madness is officially underway, and everybody is hoping their bracket will win out. But what if it turned out that we all lost—to social media? Is it possible Twitter actually has the best picks of them all?

This is what the data folks at Prime Visibility are suggesting. According to their research, the company has accurately predicted 74 percent of games in the last two years, based on the public's positive comments toward each team. That would have put their choices in the top 5 percent of all brackets last year. It's a better performance than simply picking the favorites according to gamblers—who win 71 percent of the time.

For example, social media's bracket on Thursday night probably beat yours. The company got 13 out of 16 games correct—the three losses (Baylor, Iowa State and LSU) were each by a single point. That puts it ahead of 98.7 percent of brackets at ESPN, and with hope to do even better: All of its Sweet 16, Elite Eight and Final Four teams are still alive.

Prime Visibility is a full-service digital marketing agency with experience in various channels including social media. The results for its March Madness bracket came by combing through Twitter and Facebook, looking for positive mentions throughout the entire college basketball season for each team. The company has analyzed 4.25 million positive mentions—ignoring neutral or negative ones, such as "Kentucky is playing a game today." Here is what its 2015 bracket looks like:

"We were perfect in predicting the Final Four last year," said David Neuman, director of social media services for Prime Visibility. Those schools were Florida, Connecticut, Wisconsin and Kentucky. Last year alone proved tricky in picking that lineup as a No. 7 and a No. 8 seed made it in—very unusual historically.

"It can certainly provide insight into which team is more likely to win" -David Neuman, Director of social media services, Prime Visibility

In 2013, the same approach would have correctly chosen three teams of the Final Four (Syracuse, Michigan and Louisville), while missing on Wichita State. This year, they have Kentucky winning it all—the most logical outcome given its undefeated season and massive publicity—with Virginia, Arizona and Duke making the Final Four. They already predicted upsets by UCLA and Ohio State.

Focusing only on the positive

Notice that the research specifically ignored negative feedback, and only focused on the positive. "If you look at what generates the most positive mentions about a team, it's things like a great win, a great play, or a player having a great game," said Neuman.

The same approach hasn't worked as accurately in predicting NFL playoff games—but there are far fewer games there, both in the playoffs and the regular season. The larger sample size of the college basketball pool shows that when more data are available, the wisdom of the crowds shines through. And one important wrinkle is that the research considered the full season of data—not just the last couple weeks, when fans can get carried away with "hot streaks" and "momentum."

Overall, the research suggests that some of the so-called expert wisdom might be more simply ignored in favor of what the broader world thinks, experts and nonexperts alike. Neuman points out that "while social media mentions aren't 100 percent accurate in predicting outcomes of a game, it can certainly provide insight into which team is more likely to win, as it factors in all of the positive things a team has done throughout the year." In just a couple weeks, we'll know if the public was just as accurate this year as in the past.