The Dance-Card Model Picked 35 Of 36 At-Large Bids This Year

The Dance-Card Model Picked 35 Of 36 At-Large Bids This Year

The bracket-predicting model developed by Jay Coleman, Mike DuMond, and Allen Lynch had another strong performance in 2014, and has now predicted 108 of 110 at-larges over the last three seasons. The only miss was N.C. State, which the model put outside of the bubble, in favor of California.

This is a somewhat notable miss. Based on data from recent bracket selections, the committee has shown a small bias in favor of borderline squads in the Pac-12 , which the Dance-Card has attempted to account for. According to Professor Coleman, if this model adjustment were removed for this year, California would have fallen just barely below the bubble line, although Southern Miss, not N.C. State, would have moved up.

1
1
Pending approvalOriginal post by Reuben Fischer-Baum on Regressing

This Model Has Predicted 73 Of 74 NCAA Tournament At-Large Bids

Part of the fun of March Madness is obsessively guessing who'll fall on either side of the bubble, but a model known as the "dance card"—which has predicted 73 out of 74 at-large bids over the last two seasons—might be spoiling the surprise.

First developed by Jay Coleman, Mike DuMond, and Allen Lynch back in the late 90s (although it's gone through a couple iterations since then), the "dance card" predicts which teams will get selected as an at-large using the same basic data points put in front of the committee. Many of the most predictive variables are pretty obvious (the committee cares about RPI and wins against the top 25), but it does turn out that teams take a major penalty for every game they are below .500 in road and conference play, much larger than the boost they get for each game they are above .500. Not good news for bubble teams like Minnesota, which sits at just 8-10 in conference and 4-7 on the road.

You can check out the full rankings here. While predicting 73 out of 74 at-large bids over the last two seasons is pretty remarkable, it's worth mentioning that historically the model hasn't performed quite as accurately.

Below are the dozen or so teams on either side of the "bubble", through yesterday's games. From left to right, the figures in the table are dance card rank, dance card score, chance of at-large IF today was Selection Sunday, and RPI (old formula).

This Model Has Predicted 73 Of 74 NCAA Tournament At-Large Bids

The bubble line is dependent on the number of top teams that win auto-bids, so the most useful figure is the relative rankings of the marginal teams. If you'd checked yesterday, Nebraska would have been at position 48, but a huge home win over #9 Wisconsin has bumped the Cornhuskers up seven spots and put them on track for their first NCAA tournament appearance since 1998.

[Dance Card]

1