March Madness: Hack your NCAA bracket
The NCAA selection committee has announced the 68 teams that will be competing in this year's "March Madness" national championship tournament, which kicks off (with the play-in games) on Tuesday. Much of the nation is now considering the advice of college sports pundits as well as their loyalties as they puzzle over their bracket and seek an edge in their NCAA pool.
We're here to let you in on a little secret: You can improve your odds and pick the likeliest upsets without ever having seen any of the teams in action. (In fact, having watched these teams play might even be counterproductive.) Doing so involves using the advanced statistics that have revolutionized baseball and other sports and are now coming into their own in college basketball.
But first, a little background. The selection committee uses a metric called the Rating Percentage Index, or RPI, to decide who gets into the tournament. It is not the only factor: The NCAA says RPI is just "one of many factors used by the committee." But as the Wall Street Journal has documented, a team's odds for getting into the tournament have been closely correlated to their RPI. RPI may be one of many factors the committee considers, but it seems to be a pretty important one.
Yet according to Ken Pomeroy, who runs a well-respected college basketball rankings site, the RPI is behind the times. RPI takes into account three factors: A team's winning percentage, which accounts for 25 percent of the RPI; the strength of schedule of their opponents, which accounts for 50 percent; and their opponents' opponents strength of schedule, which accounts for the last 25 percent. RPI also weights away wins so they count more than home wins, and away losses count less than home losses.
One big problem with this system is that it does not take into account margin of victory - a 40-point win counts the same as a one-point win. The NCAA has a reason not to take into account margin of victory, since they don't want teams running up the score. But it means that teams that have won a lot of close games end up with a higher RPI than they probably should, while teams with a lot of blowout wins have a lower RPI.
Another problem is the big focus on strength of schedule. A team that plays a tough schedule can lose a lot of games and still keep a relatively high RPI; a team that plays a weaker schedule can win almost everything and end up with a relatively low RPI. Critics say the system discriminates against teams from weaker conferences, and complain that which teams you play get undue consideration compared to whether you actually win or lose.
"The RPI is pretty outdated," says Pomeroy, whose statistical analysis has become an important tool for many collegiate coaches. "It was developed over 30 years ago. There are better systems out there."
Pomeroy's system is predicated on offensive efficiency and defensive efficiency, with adjustments for various factors. It considers, essentially, how often you score on each possession, and how often the other team scores on you. Pomeroy acknowledges his system isn't perfect - last January, he responded to critics who pointed out that Wisconsin occupied the #2 spot on his rankings despite losing five of 17 games. But he says that along with other more advanced ranking system, it provides insights into where the RPI may be overvaluing or undervaluing a team.
"You shouldn't rely on my system solely," he said. "But if there's a consensus among systems that a team is really underrated, that's a good guide for picking an upset in your bracket." Other advanced ranking systems include ESPN's BPI system, Massey ratings and Sagarin ratings. All these systems include considerations they think help refine their accuracy - ESPN's BPI, for example, accounts for whether a team is missing one of its key players in a win or loss.
Let's look at how this plays out this year. Pomeroy, for example, lists Pittsburgh as the #8 team in the country. Sagarin has the Panthers at #10, and Massey has them in the top 20. Yet the Panthers have been assigned an 8 seed in the West, far lower than the advanced ratings suggest they deserve. That means they are both a good pick to win their first round game and not a bad candidate to take out the region's top seed, Gonzaga, in the second round. And look at Florida: Pomeroy has them as the top team in the country, yet they are just a 3 seed in the South. The discrepancy can be explained by the fact that the Gators have lost all six of the single-digit contests in which they've been involved.
In the end, of course, any prediction is nothing more than an educated guess - which is part of the reason the tournament is so much fun. The limitations of predictions in college basketball are exacerbated because there is less of a record for statisticians to work with, says sportswriter Will Leitch.
"Much of the issue to is that, well, we're talking about teenagers," he said. "In baseball sabermetrics, you're talking about players who have been playing for years and years and years and have honed themselves into something approaching predictability. College kids are rather notoriously less reliable."
There's also another, easier (if arguably less fun) technique for improving your chances, Leitch points out: You can simply take a look at the Las Vegas odds, which reflect the collective wisdom of the crowd. (Some bettors, its worth noting, use Pomeroy's analysis to guide their bets.) History, he said, suggests the "5-12" matchup that Vegas shows with the closest point spread is the one most likely to reward an upset pick.
Of course, your strategy for winning your pool should also depend on who you're going up against. If your opponents will be expected to pour over the stats gurus, pundits and Vegas odds, it might be smart to resist the trendy upset picks. If you expect that your opponents will pick a #1 seed to win it all, you might want to consider a #2 or #3 seed. And try not too be guided too much by your own biases: You may love your alma mater, but it they're a #14 seed you probably don't want to pick them to make the Final Four.