Visualizing the Final Four (with a special bonus!)
As I've mentioned before, I also specialized in information visualization in addition to machine learning during my grad school work at Georgia Tech. I added sparklines to the main site last season, and last week I posted net efficiency margin (NEM) charts for one of the Sweet 16 previews.
Today, I'm going all out. Here are NEM charts for each of the Final Four teams updated through last weekend's games. The blue line is offensive NEM (higher is better) and the red line is defensive NEM (lower is better). The larger the gap between the two lines, the better the total NEM and thus the better the overall performance. Stay tuned to the end for an extra-special bonus!
Connecticut was consistent on offense and improving on defense until their February 11 home win over Syracuse (game 24), when they lost Jerome Dyson. When I wrote about this before Selection Sunday, the Huskies were showing signs of struggling on offense. The chart shows steady improvement in the offense since then, but a also a dip (more a return to earth after their brief excursion in -30 NEM territory) in their defense. What the chart doesn't show (because it's a five-game moving average) is a decrease in total NEM over each of their first four tournament games. After pasting Chattanooga (63.0 NEM) and Texas A&M (56.2 NEM), they "only" put up 40.2 against Purdue and 38.0 over Missouri. (The 39.1 NEM average over those two games is still well above their season performance, so there's no cause for concern.)
Michigan State dealt with injuries for much of the early part of the season, but got the offense going in mid-January. After their February 4 home win over Minnesota (game 22), the offense went into a bit of a funk. Fortunately, the defense also improved dramatically around the same time. Unlike UConn, Sparty has shown improved performance in each of their four NCAA tournament games, capped off by a 50.6 NEM performance in their win over Louisville in Indianapolis.
North Carolina has been nothing if not steady on offense all season, as I pointed out in my preview of the UNC-Villanova game. While the Heels' defense is often underrated because of their pace, an ominous trend is obvious starting after their January 21 home win over Clemson (game 18). As of their ACC Tournament loss to Florida State, their defensive NEM was approaching 0; they were essentially playing defense at the level of an average Division I team. They must have "flipped the switch" once the bright lights went on, because they've averaged -19.7 defensive NEM in their four tournament games.
Villanova was consistently playing good defense through their February 1 home win over Cincinnati. Over the next seven games they played at a -1.3 NEM average level, before recovering to around the same level as before. Fortunately during that bad stretch, the offense picked it up and the Wildcats only lost one game (at West Virginia). The defense has improved since the February 19 low point against Rutgers (9.7 defensive NEM, or 9.7 points per 100 possessions worse than an average Division I team), capped off with the -44.2 NEM against Duke in the Sweet 16.
Probably the most amazing thing about Florida's back-to-back championships is how different they were. In 2006, a young team came into their own and turned on the defense in the NCAA Tournament. In 2007, the defense was somewhat better on average than in 2006, but there was no 2006-like defensive surge. Instead, a more experienced Gator team averaged an amazing 29.5 offensive NEM over their last ten games following the late-season slump when they lost three of four. None of the Final Four teams this year has shown a tournament surge like either of these Florida teams.