Over at ESPN.com, I have submitted an entry in their Tournament Challenge contest. In this contest, you must predict the winner for each of the 63 games in the NCAA Men's Division I Basketball Tournament and then predict the final score in the championship game.The entry that simply picked the team with the higher Rogers Poll rank to win each game went 23-9 in the first round and 14-2 in the second round for a 37-11 ledger heading into the Sweet Sixteen. In the round of sixteen, it was 4-2 in games where both expected teams met and 2-0 in games where only one of the expected teams played. In all, it was 6-2 for a 43-13 (76.8%) record heading into the Elite Eight.
My other entry, which you can see by clicking here, used score predictions that are automated based on average performances for each team. Relative adjustments are made for the difference between schedule strength and current opponent strength (per the RogersPoll.com method).
In the Sweet Sixteen round of the tournament, it posted a 3-5 record (2-1 when both expected teams played, 1-3 when only one of the expected teams played, and 0-1 in the game where neither projected team made it). The complete ledger for the first three rounds is 39-17 (69.6%). Only 1 of the projected Final Four remains. As stated previously, it is becoming obvious that this model is far more applicable to the college football domain.
Though the bracket had to project the winners of each first round game and then the winners of all projected match-ups throughout the tournament prior to the first tip off, I have continued to apply the algorithm to the actual match-ups to see how it performs throughout the tournament. After three rounds, it is 43-13 (76.8%) having gone 6-2 in the Sweet Sixteen.
Of the 13 incorrect picks thus far, 3 of them were in games projected to be decided by a single point. Another was projected to be decided by just 2 points, one was projected to be a 5-point game, and two others were projected to be 7-point deltas.
To date, I have successfully predicted the exact score of BYU (1st round), North Dakota State (1st round), and Arizona State (2nd round). I have been within 5 points for 37 teams (33.0%), and within 10 points for 69 teams (61.6%). I've been within 20 points for 99 teams (88.4%).
I have predicted the exact spread for the Washington/Mississippi State game, the Michigan State/USC game, and the Purdue/Washington game. I have been within 5 points for 13 games (23.2%), and within 10 points for 26 games (46.4%). I was within 20 points for 46 games (82.1%).
Also See:
03/25: Score Predictions: Sweet Sixteen
03/23: Bracket Predictions Update: Round 2
03/21: Score Predictions: Second Round
03/21: Bracket Predictions Update: Day 2
03/20: Score Predictions: First Round, Day 2
03/20: Bracket Predictions Update: Day 1
03/18: Score Predictions: First Round, Day 1
03/18: Bracket Predictions
