Jarrett Karnibad

Before tomorrow’s Final Four games, I wanted to provide an update on how my models have performed so far. The link to see what each model predicted in every game is below.

CLICK HERE FOR THE GAME-BY-GAME RECAP FOR EACH MODEL

Model Tracking Format:

There are two main ways to track a model’s success: bracket points and total wins.

• The bracket points format that I use is the ESPN standard format: 10 points for 1st round games, 20 points for 2nd round, 40 points for the Sweet 16 games, 80 for the Elite Eight, 160 for Final Four, and 320 for picking the correct champion. It’s important to track a model’s success using this format because the end goal is to create a perfect bracket.
• The way I track total wins is by running each game through each model, regardless of whether the matchup was originally predicted to occur in its bracket. A good example of this is the Nebraska-Iowa game in the Sweet 16. The best performing model in this format is AFT. AFT’s bracket had Florida vs Vanderbilt playing against each other in this game, but in order to track each model on the same sample size (63 games or 67 games with First Four included), I have AFT, and all models, predict the matchup that actually occurred. In this case, it was Nebraska vs Iowa.

Takeaways:

The best performing model in the bracket format is FourPoint Def. It currently sits at 900 points (97.1st percentile in ESPN). Unfortunately for this model, it predicted Duke to be the champion. Had Duke been able to close out a 19 point lead over Connecticut, this model would have predicted all Final Four teams correctly. If you want to read about Duke’s late-game collapse, you can click here.

The best performing model in the total wins format is AFT. It currently has correctly predicted 50 of 60 tournament games (54 out of 64 if you include the First Four). I believe this is already a record for any of my models, and with three games to play, it has a chance to get to 53 wins. AFT predicts that Illinois and Michigan will play for the championship, and Michigan will cut down the nets.

Only six models still have their champion remaining. Three have Michigan winning (AFT, AFT Matchup, and 4 Way Game Score), and three have Arizona winning (FACTOR, kWins tSOR, and ZSCORE Sum).

The Connecticut victory over Duke was tied for the biggest upset of the tournament according to my models. Only 1 out of 24 models predicted Connecticut to beat Duke. The other biggest upset of the tournament according to my models was Texas A&M beating Saint Mary’s College in the First Round. Interestingly, there has not been a game this tournament in which no models correctly predicted the winner.

There were 20 games this tournament in which all 24 models unanimously predicted a team to win. Unanimous teams were 20-0. The average margin of victory in these games was 23.1 points. The closest game was Duke vs Siena (71-65), and the biggest blowout, unsurprisingly, was Florida vs Prairie View A&M (114-55).

Predictions:

• If you want to see exactly who each model predicted, click the link at the top of this article.
• Illinois vs Connecticut: 15 models predict Illinois, and 9 models predict Connecticut.
• Arizona vs Michigan: 13 models predict Michigan, and 11 models predict Arizona.