With just a few hours to go until the NCAA selection committee announces its tournament picks, it's time to start thinking not just about who is going to make it to the Big Dance but about what matchups they might face. That makes this the perfect opportunity to introduce an exciting new idea fresh from the spreadsheets bubbling at Giant Killers Central: the concept of Giant and Killer families.
As longtime readers know well, our Giant Killers statistical model predicts big NCAA upsets in three steps. We start with basic team power rankings, based on margin of victory and strength of schedule. Then we use regression analysis to isolate the "special sauce" variables that lead some teams to over- or underperform in Giant vs. Killer matchups. Then we apply those factors to current teams to see how much, beyond their basic strength, they resemble past Giants and Killers. (Here's a more detailed explanation.)
So far, so good. But what about style matchups? Our model rates Stephen F. Austin as a better Killer than Harvard, but couldn't there be specific opponents that find the Crimson more dangerous than the Lumberjacks? Now we can start answering that question, thanks to a data-mining technique called cluster analysis, conducted for us by Liz Bouzarth, John Harris and Kevin Hutson of Furman University.
Cluster analysis figures out how to sift through a bunch of items so that things that are similar to each other end up grouped together. Imagine a pizza with toppings that seem randomly strewn about. Cluster analysis can tell us whether some areas are heavy on pepperoni or mushroom and how to slice the pie to find those regions. Now think about a set of basketball teams; cluster analysis can tell us which statistical similarities unite various groups of teams. You already know that, say, Iowa State plays a very different brand of hoops from San Diego State. Now we have the statistical tools to relate their stylistic contrasts to how they are likely to play against particular Killers.
It turns out we can group Giants and Killers into four families apiece. We will run through them briefly then turn to matchup implications.
The four families of Giants
Giants are heavily defined by their rebounding tendencies. One family is the Roy Williams Giants, which pile up good-to-massive offensive rebounding percentages, usually with very tall players, average defensive rebounding and few forced turnovers. North Carolina, naturally, is a Roy Williams Giant this season, as was Gonzaga last season and West Virginia in recent campaigns.
A second class, the Jim Boeheim Giants, also posts strong offensive rebounding percentages, but that's because they essentially give up on defensive rebounding to focus on the offensive glass, perimeter defense and generating turnovers. In addition to Syracuse, Duke and Louisville often fall into this group. In contrast, the Bo Ryan Giants are strong on the defensive glass, are usually just average at offensive rebounding and don't generate a lot of turnovers. Along with Wisconsin, Arizona has played in this style in recent seasons.
A fourth group of Giants is basically everyone else, the teams that win enough games to earn high seeds every season without dominating any of these statistical categories. They are weak on the offensive glass without being strong at defensive rebounding or generating turnovers, and we will call them Generic Giants.
The four families of Killers
The Steph Curry Killers are sharpshooters, the bomb squads that probably come to mind first when you think of teams that have slayed Giants. (Curry scored 158 points and shot 44 percent on 3-pointers in four tournament games for Davidson, a No. 10 seed, in 2008, including wins over Georgetown, a No. 2, and Wisconsin, a No. 3.) They are far above average in the percentage of shots they take from behind the arc, tend to play at a slow pace and usually aren't so good at offensive rebounding or limiting opponent shooting. Notable Curry Killers include Cornell and Murray State in 2010, Ohio in 2012 and La Salle in 2013.
In contrast, the Kenneth Faried Killers focus on offensive rebounding. (Faried led Morehead State to two NCAA bids, including a memorable upset of Louisville in 2011, when he led the nation in OR percentage and DR percentage.) They stay in games by generating extra possessions and are usually above average at forcing steals while tending to be far below average at shooting 3s. Old Dominion in 2010 and Minnesota last season are recent archetypes.
Then there are the Jim Larranaga Killers, whose calling card is defense. Based on his studies of teams that make the NCAA tournament, Larranaga’s fundamental goal is to hold opponents to shooting under 40 percent, and under 30 percent on 3s, and this family follows the patterns of his George Mason Killers. They usually don't shoot many 3s or generate many steals. And they tend to play at a crawl, extending possessions by avoiding turnovers; their slow pace and ball protection both help them as Killers.
There is a catch-all family for Killers too, teams that don't stand out at shooting 3s, grabbing offensive rebounds, forcing steals or stopping opponent shooting. We'll call them the Generic Killers.
Some teams fit these definitions better than others, and some jump from one class to another in different seasons. But all this is actually easier to understand intuitively than to study statistically. It's pretty clear just from watching them that Manhattan and Mercer have different styles. But here's the intriguing news: Our analysis has discovered that when Giants and Killers face off, it's not just basic quality and special sauces that determine which team wins. Families matter too.
Why the specific matchups matter
Take the Roy Williams Giants, the dominant offensive rebounders. Overall, these teams are the best class at avoiding upsets. They have lost to Killers in just 11.3 percent of tournament games since 2007. But that proportion rises to 30 percent against the high-intensity defenses of the Jim Larranaga Killers. Think of Gonzaga falling to Wichita State last year or Villanova losing to Saint Mary's in 2010.
This year, North Carolina, not a particularly vulnerable Giant, would face a nightmare opponent in Harvard. Our updated model says the Crimson would have a better than 40 percent chance of knocking off the Tar Heels, thanks partly to their Giant and Killer families.
The Jim Boeheim Giants are even more prone to getting tripped up by slow-tempo defensive specialists that shut down the perimeter. This clash of styles has generated upsets in 40 percent of matchups, including Florida State's loss to Wisconsin in 2009 and Duke's destruction at the smothering hands of West Virginia in 2008. Meanwhile, Boeheim Giants have lost just six of 45 games (13.3 percent) to other classes of Killers. Lesson for 2014: Watch out, VCU. As a Giant this year, the Rams could easily fall into a 6-11 game against Tennessee. The Vols' monstrous offensive rebounding, defensive efficiency and slow pace are just the combination that causes problems for the Rams' style. Matchup factors would actually make Tennessee the favorite in this game, according to our model.
In contrast, Generic Giants, who aren't particularly strong on the glass at either end, want to face grandma-paced grinders. History's brackets are strewn with relatively weak Goliaths felled by snipers or high-possession Killers. So if you're Oklahoma, in a 5-12 matchup, you would much rather take your chances with Dayton (upset odds: about one in four) than North Dakota State (upset odds: more than one in three). Even though the Flyers are a better team than the Bison, history says they are less threatening to a squad like the Sooners.
The Bo Ryan Giants, who pile up defensive rebounds, don't like to face the Kenneth Faried Killers, who go after offensive rebounds; 23.8 percent of these faceoffs have ended in upsets while Ryan Giants have lost just 12 percent (9 of 75) of their other tournament games to Killers. Remember that last year Wisconsin went down to Ole Miss, a team that was famous for Marshall Henderson but whose offensive efficiency was based on adding possessions by grabbing boards and avoiding turnovers. The Giant to watch here is San Diego State, the weakest of the Ryans, though there aren't many Killers that fit the Faried family profile in 2014. If an above-average offensive-rebounding squad like SMU or Xavier gets a crack at the Aztecs, our model pegs the chances of an upset at more than 30 percent.
Finally, we would like to explore the exciting possibilities of sharpshooting Killers picking off Giants, but there just aren't many this year. Only one member of the Steph Curry family has a Killer rating above 10: Eastern Kentucky, a likely 14- or 15-seed. For the Colonels to have any shot at making history and justifying our model's love, they need to face a Generic Giant, not a heavy-rebounding team. Steph Curry Killers have slain Generic Giants in 46.2 percent of their matchups, and have actually outscored those Goliaths overall, while winning just 15.3 percent (9 of 59) of their other games vs. Giants. Eastern Kentucky wants Iowa State, against whom its upset odds (19.7 percent) are three to four times greater than against Kansas, Duke or Villanova.
Even 19.7 percent is a pretty slim chance. But our new family trees make it even clearer that this is a lean season for Killers. We'll have much more on that as matchups move from the hypothetical to the real on Sunday night.