About this time last year, I wrote about the factors that can the Crashing the Dance selection and seeding models can't account for. I'll probably touch on these again before next Sunday, but in the meantime I wanted to launch a new daily feature: BubbleGrid.
BubbleGrid is a table of the teams that are still on CTD's bubble, which essentially means all teams in the "Don't Get Comfortable" section of the selection report. The primary attributes considered by the selection model (plus a few extras) are listed for each team with "good" and "bad" attributes highlighted in green and red, respectively.
What do "good" and "bad" mean? I went back through the nine years of training data and determined the values for each profile attribute that a very high percentage of the selected (good) and non-selected (bad) teams had for that attribute. In other words, they are the thresholds for each attribute that might not guarantee at-large selection (or being left out), but give us a good idea based on the committee's past selections. Values above the good threshold are highlighted green, and values below the bad threshold are highlighted red. The more green and the less red, the better
Let's look at Penn State as an example. Their RPI Top 25 and Top 50 wins are highlighted green. This means that, in the previous nine seasons, teams with 3 or more Top 25 wins or 6 Top 50 wins (not necessarily together) were very likely to be make the field at-large (or were automatic qualifiers that would have been selected at-large). However, their RPI Top 200 +/- is highlighted red. This means that a very high percentage of teams with a +/- of 2 or less were left out. (As with our selection model, I only consider the RPI Top 105 teams to measure the "high percentage" of teams. Looking at the bottom 200 teams doesn't really tell us much, since the committee doesn't consider them.)
Obviously, the problem is that most bubble teams in any year have more blemishes than a Proactiv commercial. Most bracketologists guess (whether educated or otherwise) which parts of these teams' profiles are more important than the others. Because we're talking about 10-ish profile attributes for 30-ish teams, BubbleGrid makes it easier to see at a glance how these teams compare. It also allows us to quickly identify teams that the selection model might be handling badly. (Exhibit A: Georgetown)
I plan to update BubbleGrid daily, and I hope you find it useful. As always, feel free to fire off your comments and questions.
- Hover over a value for more information
- Shaded teams are predicted to be in the field
- Teams with a * are the projected automatic qualifier (AQ) for their conference
- Teams are listed in order of their selection confidence by the CTD model. In other words, the last shaded non-AQ team is the last team in, and the first non-shaded team is the first team out.
|1-25||1-50||1-100||1-200||Road||Good road wins||Bad home losses|
|Ohio St.||19-9||1||5||-3||8||-1||3||4-6||#45 Michigan, #51 Miami FL|
|Wisconsin||17-11||1||4||-5||9||-1||2||4-7||#45 Michigan, #59 Penn St.|
#62 Virginia Tech
|Boston College||20-10||3||4||0||8||0||6||5-5||#3 UNC, #56 Maryland||#260 Harvard|
|Penn St.||21-9||3||6||-3||7||-2||2||5-5||#6 Michigan St., #21 Illinois|
|UNLV||21-8||4||4||1||9||3||4||6-5||#9 Louisville, #19 BYU|
|Georgetown||15-13||4||4||-4||7||-4||0||3-7||#4 Connecticut, #13 Villanova|
|Maryland||18-11||2||3||-5||8||-2||2||2-5||#135 Morgan St.|
|Miami FL||16-11||2||2||-5||6||-4||0||4-6||#58 Boston College|
|San Diego St.||18-8||1||2||-3||4||-4||4||5-5||#50 UNLV|
|New Mexico||20-10||2||4||-1||6||-1||6||5-6||#148 Central Florida|
|Northwestern||17-11||2||6||-2||6||-4||-2||4-7||#6 Michigan St., #36 Purdue|
|Kansas St.||19-10||1||3||-2||5||-3||0||5-5||#33 Texas A&M, #34 Texas|
#75 Cleveland St.
|South Carolina||20-8||0||0||-4||7||0||4||4-6||#66 Baylor|
|Kentucky||19-11||3||3||-2||7||-2||0||4-5||#17 Tennessee||#125 VMI, #186 Georgia|
|Illinois St.||23-8||0||1||0||6||1||10||7-6||#206 Indiana St.|