Andy Cox, the man behind Crashing the Dance, agreed to answer a couple questions about Michigan’s tournament chances and what he does over on his website. I’m a big fan of Crashing the Dance because it is a new take on bracketology and the model seems to work quite well. CTD got 32 of 34 at large teams correct and seeded 82% of them within 1 seed of their actual spot on the bracket. There are also plenty of other cool features like sparklines (Michigan: ) which detail a team’s season in just a few pixels or the bubble grid which provides a daily rundown of every bubble team’s measurables. Thanks again to Andy for taking the time to answer these questions.
For people unfamiliar with the site, how exactly does Crashing the Dance work?
Glad you asked. Crashing the Dance learns from past brackets how the NCAA selection committee selects and seeds teams with certain kinds of profiles. Like the real committee, CTD handles at-large selection and seeding separately. First, it predicts whether each team would be in or out and how confident that prediction is by examining teams with similar profiles from previous years and whether those teams were selected (or conference champions that would have been selected). The top 34 at-large teams are chosen after ranking the at-large candidates by the prediction confidence. After determining the 65 teams in the field (34 at-large plus 31 conference champions), each team is assigned a most likely seed by again comparing its profile with similar teams (and their seeds) from past brackets. The full 1 to 65 seed list (called the S-Curve) is generated by ranking the teams by this most likely seed.
Let’s look at Michigan (through Monday’s games) as an example for each step. CTD thinks Michigan will be selected at large with 99.9% confidence based on how teams with similar profiles were handled in previous years. Michigan ranks 29th in at-large selection confidence out of all at-large candidates, so they are the sixth to last at-large team selected. After putting the Wolverines in the field, CTD gives them 8.80 seed points (16 seed points = #1 seed, 1 seed point = #16 seed), again based on past comps. In other words, teams with similar profiles in the past have generally been placed somewhere between a #8 and #9 seed (slightly closer to #8 here). When we rank all 65 teams by their seed points, this puts Michigan at #37 on the S-Curve, which results in a #10 seed. Note that the most likely seed is not always the same as the final seed, because more than four teams on each seed line could be given the same most likely seed. This is true here, as 7 teams have between 9.23 and 8.80 seed points (i.e., the model’s best guess is about an #8 seed). A tight cluster like this also means big jumps are possible.
What do you think of Michigan’s resume right now? The question on everybody’s mind is whether they need to beat Iowa in round 1 of the Big Ten Tournament.
I wouldn’t say they’re out if they lose, but it’s a lot like dealing with bad calls by officials. Bad calls happen, but if you take care of business the bad calls won’t make the difference between winning and losing. It’s the same thing with the selection committee. Win as many games as you can to make their decision easy. Too much can still happen this week for this game to be the decider, but the win at Minnesota certainly helped.
Also, Butler’s loss tonight kills one more available bubble slot, so the Iowa game becomes more important. I’m still not willing to say it’s a must win, because that depends on what the other bubble teams do, but as I said take care of business and don’t leave it up to others.
What are the strengths and weaknesses of U-M’s resume? We talk a lot about the top 50 wins and the road record, are there other subtle attributes that we need to pay attention to?
Top 50 wins are definitely important, and “good” road wins are probably as important as overall road record. Last year, the committee chair said that Oregon was selected in large part because of three “really good” RPI Top 100 road wins. Granted, one of those those was against #92 Cal, but the committee does like road performance. Also, going 3-1 against Minnesota and Penn State can’t hurt if it comes down to a head to head discussion.
How many teams do you see making the field from the Big Ten? By my untrained eye I feel like there are eight teams that have resumes that look “tournament worthy”.
Much of the criticism I’ve heard of the Big 10 is that there are a lot of good but not great teams (seven in the RPI Top 50), somehow inflating the profiles of mediocre teams (seven teams finished between 11-7 and 9-9) with all of the opportunities for top 50 wins. Of course, part of the reason for so many top 50 teams is good non-conference performance, at least according to the RPI’s standards. That said, while there are eight teams in now, the conference tourney is sure to cause some attrition and at least one will probably fall out. That’s another good reason why it would be good for Michigan to take care of business against Iowa.
Right now you have the Big Ten bubble teams ranked in this order: Ohio State, Minnesota, Wisconsin, Michigan, Penn State. Is Minnesota really ahead of Michigan despite the fact that Michigan has more top 50 wins, a better out of conference resume, the same in conference record, and swept Minnesota?
Through Monday’s games, Minnesota, Wisconsin and Michigan are ranked 35-37 on the S-Curve separated by 0.24 seed points. A difference this small can change quickly, often by things outside a team’s control (e.g. an opponent dropping out of the RPI Top 50 which affects the number of Top 50 wins). Unfortunately, we don’t consider head-to-head games, but I suspect the committee only tends to use those a tiebreakers because of the sheer number of possible head-to-head comparisons over 5000+ games.
Minnesota has a slightly better non-conference RPI (partly because Michigan has four games against RPI 300+ opponents) even though Michigan has the stronger schedule. I don’t like looking only at strength of schedule (in fact, I don’t use SOS explicitly for either selection or seeding) because there has to be a balance between playing tough games and winning them. Michigan’s two top 50 non-conference wins are better than Minnesota’s one, and the loss at Maryland doesn’t look so bad after North Carolina also lost there. My guess is that to the human eye Michigan’s profile is slightly above Minnesota’s. However, as a wise man once said, all bubble teams have some warts – that’s why they’re on the bubble.