For the first time, I decided to submit predictions for the annual Kaggle March Madness prediction contest. Since I have the predictions at hand I thought I would post them as each round progresses with any comments about what it’s seeing.
The model I used is pretty simple by some standards. It relies on adjusted measures of team performance–both from Ken Pomeroy as well as my own generated from a hiearchical model. The model was trained on games (regular season and tournaments) between 2003 and 2018 (without 2018 tournament games, of course). Data gathering was made infinitely easier thanks to Samuel Firke.
In terms of the play-in games, here’s what the model thinks:
| Team 1 | Team 2 | Predicted Winner (%, odds) |
|---|---|---|
| St. Bonaventure (11) | UCLA (11) | UCLA (60%, 1.5:1) |
| LIU (16) | Radford (16) | Radford (73%, 2.7:1) |
| Syracuse (11) | Arizona State (11) | Arizona St. (52%, 1.1:1) |
| North Carolina Central (16) | Texas Southern (16) | Texas Southern (68%, 2.1:1) |
The surest game here according to the model is Radford-UCLA, with Radford a 2.7 to 1 favorite to advance. Texas Southern is roughly 2 to 1 to been NC Central. UCLA is the favorite over St. Bonaventure, but not by much. And Syracuse-Arizona State is essentially a toss-ups.
Once the tournment fully kicks off on Thursday I’ll post a link to my full slate of predictions. These will be straight probabilities from the model.