This feature appears in the March 22 issue of ESPN The Magazine.
Why do teams win? It may seem like a silly question, especially when an extreme imbalance of talent and athleticism makes it all too obvious which team will emerge victorious. But not every matchup is so easy to figure. What makes the NCAA Tournament so compelling is it allows most favorites to triumph. Most, but not all. And picking those few underdogs is, in the end, more scientific than you realize.
We've combed through bales of stats from the past four Marches for our online Giant-Killers series -- the one in which we try to predict big upsets at the Big Dance -- and you know what? We've actually learned a thing or two. Like this: The most reliable indicators of upsets are offensive and defensive efficiency, schedule strength, turnover rate and three-point shooting. They help identify not only the Davids most likely to topple Goliaths but also bracket-busters further down the line.
As a primer on how our statistical markers affect games, let's chart the scoring patterns of two hypothetical teams (see Figure 1 below). On the graph, points scored are on the horizontal axis, and the percentage of games in which a team scored a certain number of points is on the vertical. Over the course of the season, we see a pattern that peaks at an average number, sloping downward on either side to form a bell curve. In this particular case, we have Team A, which averages 70 points a game, and Team B, which scores 90. Mathematically speaking, Team A could beat Team B only in the small percentage of cases in which their curves overlap -- basically, those games when Team A has a good night and Team B a bad one.
HINT FOR 2010: SYRACUSE, GEORGETOWN AND KANSAS STATE, ALL GIVE IT UP ON MORE THAN 21% OF POSSESSIONS.
Some matchups, however, are more prone to upsets than a basic stat like points per game suggests. Let's say Team A's true talent is more like a team that scores 80 ppg, not 70. That would make its overlap with Team B -- its chances of winning -- larger. So now the relevant question becomes: How do we determine a team's true talent?
Start with tempo. Fast or slow, pace affects the raw totals we see. So when it's time to make predictions, the better bet is to look at efficiency, that is, points per possession. Efficiency often shifts the bell curves. For example, suppose Team A, our 70-point team, usually plays at a slower pace than Team B, our 90-point team. If Team A can maintain its efficiency while playing at Team B's tempo, its upset chances increase (see Figure 2 below).
Remember George Mason? Back in 2006, the Patriots shocked the college hoops world with an improbable trip to the Final Four. They averaged just 69.4 ppg (which ranked 148th in D1), but you could chalk that up mostly to a methodical pace (63.7 possessions per 40 minutes, 286th in the country). The Patriots scored 110 points per 100 possessions (49th in D1), and in their Tourney run they showed they could win playing slow (a 65-60 win against North Carolina) and fast (86-84 over UConn).
The converse is in play at the defensive end. Take a look at the current squads at UTEP and William & Mary. The Miners allow 64.6 ppg, the Tribe 64.8. In other words, on first glance, the two defenses are pretty much identical. But UTEP's up-tempo offensive style generates nearly eight possessions a game more than William & Mary's, so on a per-possession basis, UTEP's defense is the 18th-best in the country, allowing 88.7 points per 100 possessions. William & Mary, however, is 183rd, giving up 101.5 points per 100 possessions.
Pace is only part of the equation, though. Our research indicates strength of schedule is another strong predictor of big upsets. In 2008, Davidson was a No. 10 seed, but the Wildcats had played North Carolina, Duke and UCLA early in the season. Even though they lost all three games, playing them allowed the team to enter March with the seventh-toughest nonconference schedule in the land. That Elite Eight run doesn't seem so surprising now, does it? "The purpose of that nonconference schedule was to test us, to expose us, to maybe get us knocked to the mat," coach Bob McKillop said at the time. "And in getting knocked to the mat, it challenged what kind of ability we had to handle adversity." South Alabama, on the other hand, was also a No. 10 seed that year, but its nonconference SOS was 77th in the country. The Jaguars lost to Butler by 20 in the first round.
Team A can also boost its chances of winning by increasing its scoring variability. If, for instance, it averages just 70 points a game but scores between 50 and 90 on any given night, instead of 60 to 80, its odds of winning go up. Problem is, so does the probability of its being blown out (see Figure 3 below). As pioneering statistician Dean Oliver, who now works for the Denver Nuggets, noted in his classic book Basketball on Paper, "What this means for bad teams is that if they are inconsistent they win more than they should."
There are two fundamental ways a team can go high-reward/high-risk. First, it can try to generate extra possessions by pressing and going for steals. In 2004 and again in 2005, UAB, led by twins Donell and Ronell Taylor, generated turnovers on more than 27 percent of opponent's possessions (the D1 average is roughly 20 percent). The Blazers dropped No. 1 seed Kentucky the first year (as a No. 9) and No. 6 LSU the second (as a No. 11). This year, keep an eye on Murray State and Western Carolina, which both force TOs on more than 24 percent of foes' possessions. [Ed.'s note: This story was sent to press before conference tournaments were completed.] And note that Syracuse, Kansas State and Georgetown share a prominent weakness: Each gives up the ball on more than 21 percent of possessions.
Alternatively, a team can try to maximize the value of its possessions by amping up its reliance on three-pointers. This is a risky route; shoot 10 threes and you're more likely to make one than all 10. But though you might expect all points to have an equal impact on a team's winning chances, treys are significantly more important than other types of scoring once the Tournament rolls around. Our research shows that when a low-seeded team increases its scoring by one point per game from free throws (while all other stats stay the same), its odds of beating a giant rise by 6.3 percent. But if it boosts its scoring by one point on threes, chances for an upset jump 8.1 percent. The same point is worth more because a team that shoots lots of threes raises its level of unpredictability -- the spread of its bell curve -- giving itself a better chance to beat a stronger opponent. Of course, to succeed with this strategy, you not only have to take, but also convert, chances. Takeaway for '10: Watch out for Cornell, which has six players shooting better than 37 percent from beyond the arc.
Sometimes, many of these factors converge in a single game. In last year's Horizon League championship matchup, for example, No. 17 Butler was favored by six at home against Cleveland State. During the season, though, Cleveland State had generated far more turnovers than Butler, taken better care of the ball and had even proved it was capable of slaying giants by winning at No.11 Syracuse. In the final, Cleveland State shot 52.6 percent from deep, held Butler to 21 percent on threes and committed only seven turnovers. The Vikings won, then -- of course -- went on to force 18 turnovers in a first-round upset of Wake Forest in the NCAAs.
Not every takedown will be so predictable. Then again, isn't that most of the fun?
Peter Keating is a senior writer for ESPN The Magazine.
ALARM BELLS
To understand how underdogs win, look at the bell curves below. "Frequency" is the percentage of games a team scores a given number of points. In Figure 1, Team A (the underdog) scores 70 points most often and puts up 80 as often as 60. The overlap of the curves represents when Team A beats Team B. Overlap can be increased in two ways: If Team A's subtle skills -- such as an efficient offense -- make it more prolific than raw stats suggest (Figure 2). Or if it uses high-risk tactics -- shooting more threes -- that increase chances of scoring more (or fewer) points (Figure 3). Simple, right?