Bracket Analysis: Duke, Indiana, Michigan and Gonzaga at the Top Seeds Using Wins Above Median At-Large
By Jason Lisk
The RPI is the primary tool the NCAA selection committee will use, even if it is not really the most predictive measure of NCAA tournament. If the committee just went in and blindly used RPI to choose their top teams, the #1 seeds would be Duke, Miami, New Mexico, and Florida (or Kansas, depending on whether you use the CBS RPI or ESPN RPI). I do not think anyone expects the #1 seeds to shake out that way. In fact, Joe Lunardi has only Florida and Duke in his #1 seed group, along with Indiana and Gonzaga. Jerry Palm of CBS Sports, one of the biggest proponents and experts on the RPI, has Duke, Indiana, Michigan, and Gonzaga as his number one seeds in the latest Bracketology.
You know what else has that exact combo–Duke, Indiana, Michigan, and Gonzaga–as the top seeds? The measure that I introduced last week that I am not going to call Wins Above Median At Large (WAMAL–because it rhymes with Mammal, and WAAAL isn’t cool). Setting aside the wins debate, WAMAL tries to measure teams by what a median at-large should do against their schedule. One of the problems with the RPI is how it views strength of schedule, particularly as it relates to the teams at the far end of the tournament spectrum, where it matters little if they played the 230th or 330th best team at home non-conference in their win expectancy, but it does matter a lot in the RPI.
I’ll give an example using Villanova. The last three games have been Rutgers and Marquette at home, and at Seton Hall. I estimate (rounding to nearest 5% chance of winning) that the likelihood of a median at-large team winning each of those was 90%, 60%, and 70% respectively. Add those up, and we would expect 2.2 wins for an at-large, and Villanova went 2-1.
If we do that for their entire season, Villanova is 3.3 wins below average. The Syracuse and Louisville wins have them being considered a tournament team by many. What WAMAL sees is, yes, they got one extra win there than expected. However, that is more than given away by the other twelve games against tourney contenders, where Villanova is 4-8, when the average at-large should win about 7.2 games against the same schedule. Throw in the costly loss at home to Columbia and last night’s Seton Hall loss, and that is why Villanova ranks below other bubble teams in this system.
I don’t have any delusion that Jerry Palm is looking at WAMAL when he sets his seeding. However, if you compared WAMAL and RPI, you might think he is at the top of his bracket. When you look at the top 30 teams in the RPI and compare the CBS current mock with RPI vs. WAMAL, 19 teams are seeded closer to where WAMAL would have them, while only 5 are closer to the RPI (5 others are even in both, with Connecticut not eligible).
Where the comparison actually drops off is closer to the bubble cut line. At the top, we look and say “Hey, Indiana is better than those teams ahead of them”. WAMAL agrees that they have accomplished more. Toward the end, though, RPI tends to dominate even though we recognize the flaws using the same measure at the top of seeding. That’s why Palm has four Atlantic-10 teams in the last ten in (to make a whopping seven teams from the A-10), but WAMAL looks at what they have actually done and sees Charlotte and Massachusetts wanting and well behind the cut, and Temple and La Salle very much on the line.
Here are the top 73 schools in Wins Above Median At-Large. As a reminder, this is not a predictive system, but rather an attempt at a retrodictive system that measures wins and accomplishment better than the RPI. Well, there is one exception to that. I count OT games as ties, which affects a handful of teams (Iowa State is 0-3, Kansas and Arizona State 3-0 among tourney teams, many haven’t played any). The cut line for the 37 at-larges, once we count off the automatic bids, comes in at Maryland/La Salle, about 1.9 wins below the median at-large. Five other teams come within a win of that cut line right now.
[photo via USA Today Sports Images]